hybrid solar-wind energy harvesting for embedded ...gsharma/papers/hadihybsolarwindh… · range...

fourth quarter 2017 1531-636X/17©2017Ieee Ieee cIrcuIts and systems magazIne 29

mohamadhadi habibzadeh, moeen hassanalieragh, akihiro Ishikawa, tolga soyata, and gaurav sharma

Feature

Hybrid Solar-Wind Energy Harvesting for Embedded

Applications: Supercapacitor-Based System Architectures

and Design Tradeoffs

Ima

ge

lIc

en

se

d b

y In

gr

am

Pu

blI

sh

Ing

30 Ieee cIrcuIts and systems magazIne fourth quarter 2017

1. Introductionutonomously deployed embedded systems in the medium power (1–10 W) range have recently re-ceived broad attention in the literature due to their

applications in numerous emerging technologies, includ-ing smart cities [1]–[4], environmental monitoring [5], ag-riculture [6]–[8], and emergency management [9]. These applications typically employ a network of field systems in locations with no or limited power infrastructure, requir-ing them to incorporate an autonomous ambient power harvesting solution for a seamless operation.

The appropriate energy harvesting approach for an embedded field-deployed system is primarily deter-mined by the power requirements of the target applica-tions. These requirements along with the most commonly harvested ambient power sources are categorized and compared in Table 1 and Table 2, respectively. While em-bedded systems that operate in the (1 nW–10 mW) power range can be powered from piezoelectric [21], thermal [22], microbial [23], RF [24], wind [25], and solar [26] en-ergy harvesters, medium-power embedded systems that operate in the (1–10 W) power range rely primarily on solar, wind, or a combination of solar and wind (hybrid) energy harvesters. A rich body of energy harvest-er designs exists in the literature that accepts solar-only, wind-only, or hybrid solar/wind power inputs and buffer the harvested energy in the rechargeable bat-teries [27]–[29].

Autonomous solar-only—or wind-only—field systems are susceptible to frequent power interruptions (down-time), because neither solar panels nor wind turbines can individually provide continuous power throughout

an entire day. Solutions introduced in the literature al-leviate the deficiency of these power sources by buffer-ing the surplus portion of the harvested energy in a large energy buffer such as rechargeable batteries [47], [48] or supercapacitors [49]–[51]. When there is inadequate input power to feed the embedded system (e.g., dark nights or days with no wind), the buffered energy is re-trieved to compensate for the shortage. However, if the input power is below the power consumption of the sys-tem for an extended period, the buffered energy is even-tually depleted, causing a system outage and increasing the downtime [52].

Hybrid harvesters [53]–[57] that can utilize both solar or wind power are attractive because these two sources often have complementary availability. This means that they can meet desired downtime targets without requir-ing over-provisioning of buffer and panel/turbine re-sources; an example of this is illustrated in Fig. 1, which depicts the average solar irradiation level and wind speed in the Rochester, NY region over the 2009–2010 calendar years. While the solar power availability peaks between the months of April through August, wind pow-er is the lowest during this period; outside this period, an inverse pattern is observed. In the presence of such complementarity, a hybrid harvester—powered from both solar and wind sources—is expected to provide a much steadier power output as compared to one that has a single power source.

Embedded systems powered by solar/wind harvest-ing are subject to cyclical and variable availability of power, which implies frequent charge-discharge cycles for the energy buffer used. Supercapacitors are partic-ularly attractive as an energy buffer in these systems because of their much longer lifetime under repeated charging and discharging [58]. The lower energy density

M. Habibzadeh and T. Soyata are with the Department of Electrical and Computer Engineering, SUNY Albany, Albany NY, 12203. E-mail: {hhabibzadeh, tsoyata}@albany.edu. M. Hassanalieragh and G. Sharma are with Department of Electrical and Computer Engineering, University of Rochester, Roch-ester, NY 14627. E-mail: {mhassanalieragh, gaurav.sharma}@rochester.edu. A. Ishikawa was with Department of Electrical and Computer Engineering, University of Rochester, Rochester, NY 14627. He is now with Computer Science Department, Carnegie Mellon University, PA, 15213.

Digital Object Identifier 10.1109/MCAS.2017.2757081

Date of publication: 20 November 2017

to enable off-grid deployments of autonomous systems for extended operational durations, robust energy harvesting in the medium power range (1–10 W) is essential. supercapacitor-based solar energy harvesters have emerged as a popular alternative due to their long lifetime under repeated charge-discharge cycles, low maintenance, environmental friendliness, and energy predictability and scalability. despite their advantages, such systems are not well matched with applications that require power continuously over their operational lifetime because solar power is unavailable during nights and severely reduced on cloudy days. for such applications, it is beneficial to combine solar power with another power source—such as wind—that exhibits complementary availability. In this pa-per, we present multiple solar/wind (hybrid) supercapacitor-based harvesters, leveraging existing open-source solar-only harvester designs. our designs center around three main categories that i) add wind harvesting capability to create a wind-only harvesting system, ii) use multiple harvesters for utilizing hybrid sources of power and for providing fault tolerance, or iii) use a single harvester in a time multiplexed configuration to simultaneously harvest from multiple power sources. We provide extensive experimental re-sults to document the functionality and operational performance of a representative set of these designs.

a

Abstract

fourth quarter 2017 Ieee cIrcuIts and systems magazIne 31

of supercapacitors compared with batteries, however, makes their physical size a critical issue in deployments and this has hindered the adoption of supercapacitors despite their favorable characteristics such as environ-mental-friendliness, convenient power management, energy predictability, and longer lifetime [59]. Because hybrid harvesting can reduce the required energy buff-ering capacity, supercapacitors can be immediate ben-eficiaries of hybrid solar/wind harvesters. In this paper, we propose multiple supercapacitor-based hybrid wind/solar energy harvesters.

Our designs are based on the UR-SolarCap solar-only open-source energy harvester [34], which was not originally designed to harvest wind power. As the first step of our design, we introduce the required hardware and software adaptations to convert UR-SolarCap to a wind-only harvesting system. As the next step, we pro-pose multiple topologies to implement a hybrid (wind/solar) harvesting system using one or more UR-SolarCap boards. Our last step introduces a design variant that harvests multiple power sources using a single UR-Solar-Cap; for this design, an analog time multiplexing mecha-nism is introduced that buffers and harvests individual sources in discrete time intervals, thereby reducing the component count substantially to provide a lower-cost harvester alternative.

The rest of the paper is organized as follows. In Sec-tions 2 and 3, we explain the basic characteristics of so-lar and wind power sources. In Section 4, we analyze the advantages of wind/solar hybrid harvesting using

available measurement data for assessing power avail-ability and models for harvested energy. In Section 5, we describe the general structure of energy harvest-ing systems. In Section 6, we review the open-source solar-only harvester, UR-SolarCap, and propose modifi-cations to turn it into a wind-only power harvester. The two single-source systems presented in this section are the building-blocks of our hybrid architectures. Section 7 introduces the high-level architectures and character-istics of our hybrid setups. Sections 8 and 9 provide technical details of our implementations. We present the experimental results of the hybrid systems in Section 10.

Table 1. Electric devices can be categorized into four distinct groups, based on their power consumption. This table presents the approximate power range of each category because there is no standard definition that clarifies the exact values. Medium power category is the focus of this paper.

Category Power Range Comments

Ultra low power # 1 mW Power consumption typically ranges from multiple pico Watts in standby state to a few milli Watts when fully operational. Applications include smart homes [10], surveillance [11], and environmental monitoring [12]. Authors in [13] develop a sub-μWatt in-vivo sensing node with energy harvesting capability, which can measure and report the intraocular pressure over a wireless communication link.

Low power 1 mW–1 W Example applications of low-power devices include smart transportation [14], machine interface [15], and structural health monitoring [16]. Authors in [17] develop a flexible smart sensing node that can be used in various IoT application. Each node has a power consumption of . 100 mW.

Medium power 1 W–10 W This category comprises embedded systems that have sensors with relatively high power demand (e.g., gas sensors [18] and cameras [19]). Furthermore, devices with relatively complicated pre-processing and feature extraction capabilities, along with many Internet gateways in WSNs, which provide high-rate data transmission and execute complicated routing algorithms fall into this category [20].

High power $ 10 W This category consists of devices that have traditionally been grid-connected with power consumption ranging from multiple Watts for household devices to kilo or mega Watts for industrial systems.

250

200

150

100

50

0

Sol

ar Ir

radi

atio

n (W

/m2 )

Win

d S

peed

(m

/s)

7

6

5

4

3

2Jan. Mar. May July Sep. Nov.

Month

Figure 1. average of solar irradiation vs. wind speed in the rochester, ny area between 2009 and 2010 [60]. the com-plimentary nature of the two power sources is readily evident.


Tabl

e 2.

A

com

pari

son

of p

ower

sou

rces

for e

mbe

dded

sys

tem

s. P

ower

repo

rted

in d

eciB

ell m

illiW

atts

(PdB

m) i

s co

mpu

ted

from

the

pow

er in

mill

iWat

ts (P

mW

) as

PdB

M =

10l

og10

(Pm

W)

[30]

. Inf

orm

atio

n pr

ovid

ed in

this

tabl

e is

der

ived

from

[31,

32]

, as

wel

l as

the

sour

ces

cite

d in

the

Com

men

ts c

olum

n of

this

tabl

e. T

he E

mbe

dded

Nom

inal

Pow

er (E

NP)

co

lum

n pr

ovid

es th

e ou

tput

pow

er o

f a ty

pica

l exa

mpl

e of

the

corr

espo

ndin

g po

wer

sou

rce

for a

n em

bedd

ed s

yste

m. A

mon

g th

e lis

ted

pow

er s

ourc

es, o

nly

win

d an

d so

lar

can

be m

eani

ngfu

lly s

cale

d to

pro

vide

suf

ficie

nt p

ower

for m

ediu

m-p

ower

sys

tem

s.

Pow

er

Sour

ceTy

peTy

pica

l Pow

er

Den

sity

Embe

dded

N

omin

al

Pow

erTr

ansd

ucer

Com

men

ts

Win

d M

echa

nica

l.

/mWcm

285

247

dBm

(5

0 W

)W

ind

Turb

ine

Pow

er d

ensi

ty is

dep

ende

nt o

n w

ind

spee

d. T

he li

sted

pow

er d

ensi

ty is

mea

sure

d at

a w

ind

spee

d of

6.3

m/s

[33]

. 50

W is

the

ENP

of a

five

-bla

de (3

0 cm

radi

us) w

ind

turb

ine.

Sola

r El

ectr

omag

netic

/mWcm

15

242

dBm

(1

5 W

)So

lar P

anel

s (O

utdo

ors)

(0

–200

KLu

x)

The

liste

d EN

P is

repo

rted

in [3

4]. S

olar

pan

els

can

read

ily b

e sc

aled

to p

rovi

de h

ighe

r or l

ower

po

wer

, mak

ing

them

sui

tabl

e fo

r em

bedd

ed a

pplic

atio

ns. A

35 #

33

cm2 p

anel

can

pro

vide

a ra

ted

pow

er o

f 15

W.

Ther

mal

Th

erm

al/

Wcm

15

3n

22 d

Bm

(150

mW

)Th

erm

oele

ctric

G

ener

ator

(TEG

)Th

e po

wer

den

sity

is re

port

ed fo

r a te

mpe

ratu

re d

iffer

ence

of

T10

C.cT=

Hig

her d

ensi

ties

can

be

achi

eved

by

incr

easi

ng te

mpe

ratu

re d

iffer

ence

. The

ENP

is re

port

ed in

[35]

whe

n th

e ho

t and

col

d en

d te

mpe

ratu

res

are

55Cc

. a

nd

,21

Cc re

spec

tivel

y an

d th

e TE

G si

ze is

30 # 3

4 # 3

.2 m

m3 .

Vibr

atio

n M

echa

nica

l/

Wcm

145

3n

19 d

Bm

(74

mW

)El

ectr

omag

netic

Gene

rate

d po

wer

is d

epen

dent

on

the

ampl

itude

and

freq

uenc

y of

mec

hani

cal s

timul

us a

nd

dim

ensi

ons

of th

e ge

nera

tor [

36],

[37]

. The

ENP

is re

port

ed in

[38]

for a

tran

sduc

er o

pera

ting

at

54 H

z an

d ac

cele

ratio

n of

0.5

7 g.

The

siz

e of

the

tran

sduc

er is

. 7

0 cm

3 .

Mec

hani

cal

/Wcm

330

3n

–7 d

Bm

(200

μW

)Pi

ezoe

lect

ric

mat

eria

lsPo

wer

is d

eter

min

ed b

y th

e ty

pe o

f the

pie

zoel

ectr

ic m

ater

ial a

nd th

e in

tens

ity o

f the

mec

hani

cal

stra

in [3

9]. T

he E

NP c

orre

spon

ds to

the

tran

sduc

er s

tudi

ed in

[39]

, exc

ited

with

freq

uenc

y of

12

0 H

z an

d th

e pe

ak a

pplie

d st

rain

of 2

00

.nf

The

siz

e of

the

tran

sduc

er is

13#

10#

0.38

mm

3 .

Mec

hani

cal

/Wcm

50

3n

–7 d

Bm

(200

μW

)El

ectr

osta

ticG

ener

ated

pow

er is

dep

ende

nt o

n th

e m

echa

nica

l stim

uli a

nd th

e fa

bric

atio

n pr

oces

s [4

0]. T

he

ENP

is th

e ou

tput

pow

er o

f the

sys

tem

des

crib

ed in

[40]

, at v

ibra

tion

acce

lera

tion

unde

r 1 g

. Th

e tr

ansd

ucer

is 1

cm

2 with

a m

obile

mas

s of

1 g

m.

Mic

robi

al

Bioc

hem

ical

./

Wcm

26

2n

–2 d

Bm

(600

µW

)M

icro

bial

Fue

l Ce

llPo

wer

den

sity

is d

epen

dent

on

the

area

of t

he a

node

; as

it in

crea

ses,

the

pow

er d

ensi

ty d

ecre

ases

[4

1]. F

or s

mal

ler a

node

s, th

e po

wer

den

sity

can

reac

h as

hig

h as

/

Wcm

300

2n

[42]

. The

ENP

is

repo

rted

in th

e sa

me

sour

ce fo

r a tr

ansd

ucer

with

are

a of

2 c

m2 .

Indo

or

Ligh

tsEl

ectr

omag

netic

/Wcm

15

2n

–3 d

Bm

(480

μW

)So

lar P

anel

s (In

door

s)

(1 L

ux–3

kLu

x)

Gen

erat

ed p

ower

is d

epen

dent

on

the

light

inte

nsity

and

am

bien

t tem

pera

ture

. An

indo

or s

olar

pa

nel i

s ex

pect

ed to

gen

erat

e or

ders

-of-m

agni

tude

low

er p

ower

, due

to it

s sm

alle

r pro

file

and

subs

tant

ially

low

er s

olar

irra

diat

ion.

The

ENP

is re

port

ed in

[43]

at i

llum

inan

ce o

f aro

und

1 kL

ux

and

sola

r pan

el a

rea

of 6

5 cm

2 .

Dire

cted

RF

Elec

trom

agne

tic/

mWcm

502

20 d

Bm

(100

mW

)An

tenn

aIn

dire

cted

RF

pow

er h

arve

stin

g, a

pow

er c

aste

r, lo

cate

d in

pro

xim

ity o

f the

pow

er h

arve

ster

(s) a

nd

aim

ed a

t cer

tain

dire

ctio

n, tr

ansm

its th

e RF

sig

nal,

ther

eby

impr

ovin

g th

e po

wer

den

sity

[44]

. The

EN

P is

repo

rted

in [4

5], f

or th

e tr

ansm

issi

on p

ower

of 9

10 m

W, l

ocat

ed 1

m a

way

from

the

rece

iver

.

Acou

stic

M

echa

nica

l/

Wcm

96

3n

–11

dBm

(8

0 μW

)M

icro

phon

es/

Piez

oele

ctric

Gen

erat

ed p

ower

is d

epen

dent

on

the

soun

d pr

essu

re le

vel,

dist

ance

from

the

sour

ce, a

nd

the

freq

uenc

y of

the

soun

d [4

6]. T

he E

NP is

repo

rted

in [4

6] fo

r a P

ZT tr

ansd

ucer

(.

1 c

m3 )

stim

ulat

ed b

y a

vibr

atio

n w

ith m

agni

tude

of 2

.25

m/s

2 and

freq

uenc

y of

100

Hz.

Ambi

ent R

F El

ectr

omag

netic

/nW

cm12

2

–23

dBm

(5

μW

)An

tenn

aTy

pica

l Pow

er D

ensi

ty is

hig

hly

depe

nden

t on

the

dist

ance

from

the

broa

dcas

ting

stat

ion,

type

of t

he

ante

nna,

and

the

frequ

ency

ban

d of

the

wav

e; th

e EN

P is

the

aver

age

for t

he 3

G (B

Tx) b

and

in L

ondo

n [2

4].


We conclude our paper in Section 11 by summarizing the contribution of our paper and suggesting future work.

2. Solar Power CharacteristicsSolar is typically the primary power source for autono-mous systems, mostly due to three factors:

(i) Solar power is readily available almost every-where. Even in enclosed environments, low-irradi-ation solar power transducers can be employed to generate electricity from available light [61], [62].

(ii) Solar power transducers provide higher power density compared to other sources; a two to five orders-of-magnitude advantage over RF, thermal, and airflow power sources is reported in [63].

(iii) Unlike wind turbines and vibration harvesters, solar energy can be harvested through solid-state devices with no moving parts, which translates to higher reli-ability, longer lifetime, and lower maintenance costs.

Solar power transducers are composed of elementary units called solar cells or photo-diodes that generate elec-tricity from solar irradiation. The solar cells are connected in series and/or parallel to form a solar module, (or alter-natively solar panel). Table 3 tabulates the notational sym-bols that are used to quantify the solar power sources.

2.1. Solar Cells (Photo Diodes)A solar cell is created on a semiconductor substrate, which is a bandgap material with a lower energy valence band separated from the conduction band by a forbidden energy gap of about 1 electron volts (eV). Silicon (Si) is the most commonly used semiconductor material with a bandgap ( )EG of approximately 1.1 eV at 273 K. Other semiconductor materials that are used include GaAs [64], Cds/CdTe [65], and Cu(InGa)Se2 [66].

In the normal pure crystalline state of a semiconduc-tor, electrons completely occupy the valence band and are restricted from acquiring any energy (for instance, kinetic energy of motion necessary to conduct electricity) unless they can cross the bandgap into the (normally) empty conduction band. A photon incident on the semiconduc-tor can excite an electron to transition from the valance band to the conduction band, provided that the photon’s energy Em exceeds the semiconductor band gap .EG The energy of a photon is

,E hcm

=m (1)

where h denotes the Planck’s constant, c is the speed of light in vacuum, and m is the wavelength of the photon.

Table 3. Notations used to describe solar power sources in Section 2, with equation and section references for each notation.

Notation Equations Sections Description

m 1 2.1 Wavelength of electromagnetic radiation

Em 1 2.1 Energy of a photon at wavelength m

h, c 1 2.1 Planck’s constant; . / ,h m kg s6 626 10 34 2#= - speed of light in vacuum; c 3 108#. m/s

k, T, q 2; 4; 5; 6 2.1; 2.2 Boltzmann’s constant; . ,k 1 38 10 J/K23#= - temperature in K, charge of an electron

VOC 2 2.1 Open circuit voltage of a solar cell I CS 2; 3 2.1; 2.2 Short circuit current of a solar cell, at zero terminal voltage difference I0 2; 4; 5; 6 2.1; 2.2 Diode saturation current Iph 3; 4; 5; 6 2.2 The aggregate current generated by a solar panel when exposed to light Isolar 4 2.2 The output current of a solar panel, which flows through the load Vsolar 4; 5 2.2 The output voltage of a solar panel across the terminals of load

a 4; 5; 6 2.2 Diode ideality factor; a is typically between 1 and 2 Psolar 5; 7 2.2 The amount of power a solar panel delivers to the load VMPP 6 2.2 Solar panel output voltage when operating at the Maximum Power

Point (MPP) Prated 7 2.2 Solar power measured at a certain irradiance level Wsolar 7 2.2 Instantaneous solar irradiation level (in /W m2 ) at a given point in time Wrated 7 2.2 Solar irradiation level (in /W m2 ) at which Prated is specified

PmaxSolar 7 2.2 Maximum power of a harvester imposed by its design and

implementation


In a pure semiconductor, the electrons transport-ed to the conduction band by absorption of a photon would quickly transition back into the valence band (emitting a photon in the process). In a solar cell, such recombination is avoided by creating a p-n junction diode. The substrate is doped to obtain a p-type mate-rial by the addition of a small fraction of acceptor atoms that have one less electron in their valence shell com-pared to the semiconductor. The incorporation of these acceptor atoms in the crystal creates empty acceptor energy levels just above the valence band into which electrons can readily transition to conduct electricity. An n-type material is similarly obtained by doping the semiconductor with donor atoms that have one more electron in their valence shell compared to the semi-conductor, which produces occupied donor levels just below the conduction band from which electrons can also readily transition into the conduction band to con-duct electricity. A solar cell is obtained by stacking an n-type layer on top of the p-type substrate and adding a metal layer below the p-type semiconductor and a met-al grid on top of the n-type layer. Locally, in the vicinity of the p-n junction, the electrons from the donor layers migrate over from the n-type region into the p-type re-gion till an internal electric field is set up that prevents such movement. When the solar cell is exposed to light, this internal electric field quickly moves electrons that transition to the valence band by absorbing a photon across the junction, preventing their recombination. This builds up a voltage difference between the two electrodes formed by the metal layer below the p-type substrate and the metal grid over the n-type layer. Con-nection of the electrodes to a load causes a current to flow providing electrical power from the solar energy of the photons. A detailed description of solar cell physics can be found in [67].

Short circuit current of a solar cell ( )ISC is the cur-rent flowing between the two metal layers when they

are short-circuited. The exact value of ( )ISC depends on solar irradiation and cell temperature. It is also highly dependent on the physical and chemical characteristics of the p-n junction; it is typically used as a comparison criterion to rate solar cells. Because ( )ISC is dependent on the operational condition of the cell, reference short circuit current ( )ISCR is typically used to represent the generated current at a specific temperature and solar irradiation level. It is possible to use ( )ISCR to calculate ( )ISC through either analytic or experimental equations [68]. The performance of a solar cell can also be char-acterized by measuring its open-circuit voltage ( ),VOC which represents the voltage difference between the metal layers of p- and n-type layers when they are open-circuited (i.e., measured when the solar cell current is 0 A). VOC can be calculated as:

,lnV qkT

II 1OCSC

0= +c m (2)

where k is Boltzmann’s constant, T is the temperature in Kelvin, q is the absolute value of electron charge, and I0 represents the diode’s saturation current. For a typi-cal solar cell, VOC is around 0.5 V, which is not sufficient in many applications. Higher voltage and current levels can be reached by connecting multiple solar cells in dif-ferent serial/parallel topologies.

2.2 Solar PanelsSolar panels consist of multiple solar cells connected in a series and/or parallel topology to reach higher output voltages and currents. We will use two terms, solar pan-els and solar modules, interchangeably throughout this paper. Despite the internal complexities of solar cells, solar panels can be modeled using the simple circuit shown in Fig. 2, in which the ,R1 ,R2 and R3 represent the equivalent junction, intrinsic shunt, and intrinsic se-ries resistances of the panel’s solar cells, respectively. The diode D1 represents the equivalent p-n junction of all of the solar cells, which generate a current, ,Iph based on the photons that are incident. For example, assuming N parallel-connected solar cells, Iph is:

.I N I·ph SC= (3)

Assuming a very large R2 and a relatively small ,R3 the current that flows through the load ( )Isolar can be esti-mated as [69]:

,expI I I kTq V

1·

ph 0solarsolar

a= - -c m; E (4)

where I0 is the diode saturation current, Vsolar is the panel’s output voltage, a is the diode’s ideality factor (typically between 1 and 2), and T is the junction’s

Iph

R1

R2 RL

R3

D1

+

–

VS

olar

I Sol

ar

LoadSolar Panel

Figure 2. solar panels can be modeled as a collection of photo cells with an aggregate current of Iph and a diode ( )D1 that has a non-linear intrinsic resistance ;R R1 2 and R3 are the equivalent shunt and series resistances, respectively. the actual values of the circuit elements depend on the char-acteristics of each solar cell, their quantity, and the topology of their connection [69].


temperature. Using Eq. 4, the generated power of solar panel ( )Psolar is:

.expP V I I I V kTq V

··

ph 0 0solar solar solarsolar

a= + -^ ch m (5)

As can be inferred from Eq. (5), the P-V characteris-tics of solar panels are non-linear and non-monotonic. For each solar irradiation level, the power reaches a global maximum only at a specific voltage (or equiva-lently current) value, which is the desirable operat-ing point to extract the maximum amount of power from the solar panel. This optimal operating point is referred to as the Maximum Power Point (MPP) and can be calculated by solving ,dP dV 0Solar Solar =^ h which yields the equation

,expII I

kTq V

kTq V

1· ·ph MPP MPP

0

0

a a+= +c cm m (6)

where VMPP is the solar panel output voltage when it is op-erating at the MPP. Equation (6) indicates that the MPP of a solar panel is a function of Iph and ,I0 which depends on solar irradiation ( )Wsolar and temperature ( ).T Although Eq. 6 provides a basis for accurate MPP tracking, its imple-mentation in harvesting systems is challenging because of the need to measure solar irradiation, temperature, and the voltage and current of the solar panels. Therefore, several alternative approaches, known as MPP Tracking algorithms (MPPT), have been proposed in the literature, which offer different simplicity vs. accuracy tradeoffs [69]–[75]. The power generated by a solar panel ( )Psolar at a solar irradiation of Wsolar is given by the following equation [34]:

, ,minP P WW Pmax

solar ratedrated

solarSolar= c m (7)

where the Prated is the rated power output of the solar panel at the rated irradiation level of .Wrated

3. Wind Power CharacteristicsWind turbines are used as the primary power source of embedded field devices much less frequently than solar panels because

(i) The mechanical operation of wind turbines sub-stantially increases system maintenance; even a minor damage to a blade can lead to consider-able decrease in generated power, requiring re-pair or replacement.

(ii) In contrast with solar power, for which power avail-ability changes are steady and predictable, wind power changes tend to be much more random. The uncertainties about power availability make the system design and provisioning more challenging.

(iii) Unlike solar panels, wind turbines generate an AC power output, requiring rectification circuitry for

embedded systems that typically operate using DC power.

Despite these limitations, the fact that wind power possesses near-perfect complementary characteristics to solar power (as we detail later in Section 4) makes it an attractive power source for embedded systems that require power availability throughout the day. Table 4 tabulates the notational symbols that are used to quan-tify the wind power sources.

3.1. Wind Power SourceThe kinetic energy Ewind of a mass m of air moving with a velocity of windy can be obtained from standard New-tonian mechanics as

.E m21 2

wind windy= (8)

Using Eq. 8, the available power from air flowing through a vertical cylinder with side disc area of A can be calcu-lated as:

,P R21

u2 3

windt r y= (9)

where windy is the velocity of the wind arriving at the blades of the turbine, ,A R2r= and t is the air density, which depends on altitude, air temperature, and air chem-ical composition. When the wind traverses the turbine, ki-netic energy form the air gets transferred to the turbine as rotational motion, slowing the wind to a downstream wind speed downy that is lower than the upstream wind speed

.windy The corresponding wind power decreases from the upstream wind power of Pu to the downstream wind power Pd obtained by replacing windy in Eq. 9 with .downy The difference P Pu d- between the available upstream and downstream wind power is transferred to the turbine, as shown in Fig. 3. The energy transferred to the turbine is typically modeled as [76]:

( , ) ,P C R21

p2 3

wind windb t r yK= (10)

where Cp is the turbine power coefficient, which is based on the turbine’s pitch angle b^ h and the ratio K of its blades’ circumferential speed to the velocity of upstream wind flow. For each value of upstream wind speed ,windy Eq. 10 determines an optimal value optK for K that achieves the maximum power output .Pmax To use Eq. 10 in that Maximum Power Point Tracking (MPPT) algorithms additional sensors for measur-ing the turbine’s rotational speed and upstream wind speed are required. Several MPPT approaches are in-troduced in the literature that use speed sensors or rely on use of adaptive algorithms to eliminate the need for sensors [76]–[80].


3.2. Wind Turbine TypesFor the conversion of mechanical power into electric pow-er, wind turbines used in embedded systems typically em-ploy a variable speed induction generator. A representative design for such a system is shown in Fig. 3. Wind turbines generators (WTGs) are constructed using two parts, a rotor and a stator, and operate based on the principles of elec-

tromagnetic induction. A turbine’s blades capture the me-chanical power of the air flow and transfer it to the rotor; the rotational movement of the rotor generates a changing electromagnetic field, which induces a voltage difference across the windings of the stator.

In synchronous turbines used for grid applications, the electromagnetic field within the generator is created

Table 4. Notation used to describe wind power sources in Section 3, with equation and section references for each notation.

Notation Equations Sections Description

Ewind 8 3.1 Kinetic energy of a mass of air particles flowing with the speed of windy

m 8 3.1 Mass (of a volume of air) Pu 9 3.1 Kinetic power of a mass of air flowing into a wind turbine

(upstream)

windy 8–10; 12; 13; 15–18 3.1; 3.3; 3.4 Speed of upstream wind, arriving at the blades. uwindy y= Pwind 9; 10 3.1 Power captured by a wind turbine, a portion of Put 9; 10 3.1 Air mass density

R 9; 10 3.1 Radius of circle spanned by the rotating turbine blades

( , )Cp bK 10 3.1 Wind turbine power coefficient

K 10 3.1 Ratio of blade circumferential speed (at the tip) to velocity of upstream wind flow

optK 10 3.1 Optimal ratio of blade circumferential speed (at the tip) to velocity of upstream wind flow

b 10 3.1 Turbine’s pitch angle Vs 13; 14; 15 3.3 Wind turbine stator voltage K 13; 15 3.3 A constant representing physical properties of a wind turbine z 13; 14; 15 3.3 Electromagnetic flux of the permanent magnet (in the stator) ~ 12; 13 3.3 Angular velocity of a wind turbine’s blades Vturbine 14 3.3 Wind turbine output voltage P turbine 14; 15; 16; 17 3.3; 3.4 Wind turbine output power Xs 14; 15 3.3 Wind turbine synchronous reactance RA 14; 15 3.3 Wind turbine stator winding resistance D 11; 12; 13; 14; 15; 16 3.3 Harvester duty cycle to emulate a variable load

( )R DL 11; 12; 14 3.3 Wind turbine equivalent load as a function of duty cycle Rcons 11 3.3 Wind turbine load Kturbine 16 3.3 A wind turbine-specific constant to describe the wind MPP

Dopt 16 3.3 Duty cycle at which maximum amount of power is extracted

cut outy - 17 3.4 Wind turbine cut-out speed

cut iny - 17 3.4 Wind turbine cut-in speed

ratedy 17 3.4 Wind speed at which the turbine delivers a power of Prated

P rated 17 3.4 Rated power output of a wind turbine @ ratedy ( )f windy 17; 18 3.4 Generated power of wind turbine expressed as a function of

wind speed

C1, C2, C3 18 3.4 Characteristic parameters of a wind generator determined by cut–in and cut–out speeds


by the rotor’s field circuit, which acts as a magnet and creates an electromagnetic field when a DC current is applied to it. Alternatively, the rotor can be constructed from permanent magnets, thereby eliminating the need to continuously supply a DC current. Induction genera-tors maintain their electromagnetic field by drawing reactive power from the grid. For off-grid applications, a block of capacitors is added to the generator to pro-vide reactive power. This technique is referred to as self-excitation. Although induction generators are relatively more reliable and less expensive, permanent-magnet synchronous generators are usually used in off-grid medium-power wind turbines due to their ability to drive reactive loads and their independence from grid and self-excitation capacitors.

Similar to electric generators, wind turbines gener-ate three-phase AC power. Because typical embedded systems require a constant DC voltage input, the out-put of the turbine must be rectified. A detailed explana-tion of synchronous and induction generators can be found in [81].

3.3. Permanent-Magnet WTG Equivalent CircuitIn this section, we focus our discussion on permanent magnet Wind Turbine Generators (WTGs) (which are also representative of the WTG used in our experi-ments in Section 10). Because permanent-magnet syn-chronous generators do not include a field circuit, their output voltage cannot be controlled by manipulating their field circuit current. A voltage regulator is there-fore required to produce a constant DC output from the rectified voltage.

Switching regulators are typically preferred due to their high efficiency and their ability to boost and/or reduce the voltage. Once rectified and regulated, the output power can be used to drive a load. From the wind turbine’s perspective, the voltage rectifier, switching voltage regulator, and the load appear as a single equivalent load resistance ,RL^ h which is con-trolled by the switching duty cycle (D) of the switching voltage regulator [82]. We can show that for a switch-ing regulator such as Single-Ended Primary Inductor Converter (SEPIC) and a rectifier bridge, ( )R DL can be represented as [82]:

( ) ( ) ( ),R D D R f D18 1· ·L

2

1consr= - = (11)

where Rcons is the actual load resistor and is assumed to be constant. The rotational speed ~^ h of the WTG is dependent on the load resistance, as well as the wind speed, written as:

( , ) ( , ) .f R f DL 2wind wind~ y y= = (12)

Figure 4 depicts the equivalent per phase circuit of a generic permanent-magnet synchronous generator. Vs represents the magnitude of electromagnetic force induced in the stator that depends on the rotational speed as:

( , ),V K K f D· · · ·s 2 windz ~ z y= = (13)

where z is the electromagnetic flux, which is deter-mined by the characteristics of the permanent magnet in the WTG. Xs represents the magnitude of generator synchronous reactance and RA is the winding resis-tance. The power output of the WTG is then obtained as

,V V R jX RR·s

L s A

Lturbine = + +

(14)

RL

RA

Vs

VT

urbi

ne

I Tur

bine

LoadWind Turbine

jXs~

~

Figure 4. Per phase equivalent circuit of a permanent-mag-net synchronous electric generator, similar to the one used in many medium-power off-grid wind turbines. Xs and RA rep-resent the stator synchronous reactance and its winding re-sistance, respectively. RL is the equivalent resistance seen by the wind turbine, which can be varied by manipulating the duty cycle (d) of the switching regulator that functions as a variable load.

Pd PuPturbineStatorWinding

Rotor

NS

ω

Ups

trea

m W

ind

Flo

wat

Spe

ed =

vu

= v

win

d

Dow

nstr

eam

Win

d F

low

at S

peed

= vd

Figure 3. operation of a permanent-magnet wind turbine, operating as a synchronous generator. the incoming wind at the speed of windy has a kinetic power of .Pu a portion of this power ( )Pturb ein is transferred into the turbine, causing its blades to turn at an angular velocity of ~ and lowering the departing wind speed to dy and its kinetic power to .Pd


.

( )( )

( ) ( , )

P RV V

R jX RR

V f D

K f D f D

·

·

· · ·

Ls

L s A

L

s

22

2

23

2 23 2

2

turbineturbine

windz y

= =+ +

=

=

(15)

Although, one cannot usually obtain a closed-form solution, Eq. 15 implies that the output power of the wind turbine is dependent on turbine implementation, wind speed (through rotor angular velocity ~), and the equivalent variable load .( )R DL Conceptually, the opti-mum duty cycle ( )Dopt that generates the maximum out-put power can be obtained using calculus of variations:

( ),dDdP D K f0 ·turbine

opt turbine turbine wind& y= = (16)

where Kturbine is a turbine-specific constant, y is the wind speed, and fturbine is a function that is determined by the load curve of the wind turbine. In accordance with Eq. 16, the extracted power from the wind turbine can be ma-nipulated by adjusting the duty cycle of the load based on the wind speed (i.e., the availability of the wind pow-er), assuming that the load is being emulated by a switch mode circuit such as SEPIC [34], [50].

3.4. Wind Turbine Power LimitationsThe amount of power a wind turbine can generate is fur-ther limited by its design and manufacturing properties. Typically, the turbine’s generated power Pturbine^ his rep-resented as a function of the wind speed windy as

( )

,,,

,

PfP

0

0

turbinewind

rated

wind cut in

cut in wind rated

rated wind cut out

wind cut out

1 11

#

#

$

y

y y

y y y

y y y

y y

=

-

-

-

-

Z

[

\

]]

]] (17)

where , ,cut in ratedy y- and cut outy - represent, respectively, the cut-in, rated and cut-out speeds. When the wind

speed is below ,cut iny - the blades of the generator do not turn and no power is generated. For wind speeds in the range ,cut in wind rated1 1y y y- the generated power varies with the wind speed ( ) .f windy^ h The speed ratedy in Eq. 17 represents the wind speed at which the turbine reaches its highest generated power ( )P rated and this speed is de-termined by design restrictions. For wind speeds in the range ,rated wind cut out1#y y y - the output power remains at .P rated Under extreme condition where wind cut out$y y - the wind turbine shuts down to prevent a mechani-cal damage and the power is reduced to zero. Regres-sions models are typically used to quantify ( ),f windy for instance, as a second order polynomial [84] or cubic power curve [85]. A typical second order polynomial formulation is:

( ) ,f C C C1 2 32

wind wind windy y y= + + (18)

where ,,C C1 2 and C3 are the characteristic parameters of the wind generator.

4. Hybrid Energy HarvestingAlthough Fig. 1 demonstrates a near-perfect comple-mentary nature of solar and wind power sources over a full year window, the figure lacks critical detail about the daily availability patterns of these two power sourc-es. An analysis of the complementary nature of solar/wind power sources is necessary at a daily rather than annual scale because in most embedded systems the daily variation is the main consideration in allocating the size of the energy buffer (supercapacitors in our ex-perimental setup).

Figure 5 demonstrates the hourly instantaneous wind speed in the Rochester, NY area over three randomly-selected days of Fall 2015 and Winter 2016. We selected winter days when the average generated power of solar

40

30

20

10

Win

d S

peed

(m

ph)

01 a.m. 4 a.m. 7 a.m. 10 a.m. 1 p.m.

Time (h)4 p.m. 7 p.m. 10 p.m.

20 Dec.1 Jan.10 Jan.

Figure 5. hourly wind speed (mph) patterns in rochester, ny on dec 20, 2015, Jan 1, 2016, and Jan 10, 2016 [83]. during these winter days, wind power provides a synergistic alternative to solar power due the limited availability of sunshine. the daytime/nighttime periods are indicated using a lighter/darker background, respectively. While the solar power is zero during nighttime, there is ample wind power that can avoid buffer energy depletion.


panels is relatively low compared to summer days and the system can significantly benefit from hybrid energy harvesting, thereby making the analysis conservative. The areas with blue background in Fig. 5 represent the nighttime (midnight – 8 AM), during which solar panels do not generate any power; however, ample wind power is available on all three days, especially on December 20, 2015. Although the hour-by-hour characteristics of the wind power are stochastic, we are primarily inter-ested in determining the cumulative power available during the nighttime because a typical supercapacitor block can buffer the harvested energy, largely eliminat-ing impact of hourly fluctuations.

4.1. Quantifying The Complementary Characteristics of Solar and Wind Power SourcesTo highlight and quantify the advantages of hybrid so-lar/wind energy harvesting, we use the metric of aver-age downtime, ,tdown

yr defined as the average number of hours per year for which the system shuts down due to lack of power. The average downtime can be mod-eled using historically recorded data of solar irradia-tion and wind speed at the deployment location. Solar irradiation [60] and wind speed [83] data of US cities are available as an average value over t 1D = hour in-tervals. To calculate ,tdown

yr we assume that the system is simultaneously harvesting energy from a solar panel and a wind turbine with the rated powers of P solar

rated and ,Pwind

rated respectively. The energy is buffered in a superca-pacitor block with the maximum energy storage capac-ity of ESC

max and it is delivered to a load demanding Pload Watts of power. Although the analysis is generally for a hybrid case, it can be easily adapted for a single source scenario by setting Pwind

rated or Psolarrated to zero. Table 5 tabu-

lates the notations used in this section to quantify hy-brid power sources.

In this analysis, our goal is to determine the available energy in a system on an hour-by-hour basis through the 24 hours of a day; for this, we use the notation ( )P nin to denote the average available power during the nth hour of the day. This power is related to the solar ( ( )P nsolar ) and wind power ( ( )P nwind ) generation as:

( ) ( ) ( ) .P n P n P nin solar wind= + (19)

We use the solar irradiation data available in [60] to compute the available average solar power during the nth hour of the day ( ) .P nsolar^ h We refer to Eq. 7 and as-sume a specific solar panel size to arrive at ( ) .P nsolar Similarly, we use the wind speed data from [83] and ap-ply it to Eq. 17 to determine the wind power during the nth hour of the day ( )P nwind^ h by making a similar as-sumption on the size of the wind turbine.

During the nth time interval, an average power of ( )P nin is generated from both solar and wind power sources (Eq. 19). We assume that the power consumption of the load Pload^ h is constant throughout the entire simulation period (alternative load profiles are readily incorporat-ed). Therefore, based on whether the load consumption is higher than the generated power or not, a shortage P Pin load1^ h or a surplus P Pin load2^ h in the energy oc-

curs, which in turn depletes or charges the supercapaci-tor block, respectively. We use the notation E(n) to refer to the stored and available energy in the supercapacitor block at the end of the nth interval. Furthermore, the nota-tion ( )E nAcc denotes the accumulated energy up to the nth time interval. The relationship between E(n) and ( )E nAcc can be formulated via the recursions:

( ) ( ) ( ( ) ),E n E n t P n P1 · in loadAcc D= - + - (20)

( ) [ ( ( ), ), ],min maxE n E n E0 SCmax

Acc= (21)

where ESCmax is the energy storage capacity of the super-

capacitor block, which cannot be enforced by the har-vester through an over voltage limitation mechanism [34]. In Eq. 20, tD denotes the duration of the time inter-val (1 hour), ( )P n Pin load- denotes the power surplus or shortage; therefore, their product ( ( ( ) ))t P n P· in loadD - is the accumulated energy during the nth time interval. Equation 20 implies that a device that is powered from a

Table 5. Notation used to describe hybrid power sources in Section 4, along with equation numbers they appear in.

Notation Equations Description

(P nin ) 19; 20; 22 Average total power during nth interval

( )P nsolar 19 Average Psolar during nth interval

( )P nwind 19 Average Pwind during nth interval

ESCmax 21 Max. energy capacity of

supercap. block

( )E nAcc 21; 20; 22 Accumulated energy up to nth interval

( )E n 21; 20; 22 Stored energy at end of nth interval

P load 20; 22 Power consumed by load

(t ndownD ) 22; 23 Downtime contribution of nth interval

P load 22; 20 Power consumption of the load

t downyr 23 Annual downtime


supercapacitor block will be down if ( )E n 0Acc 1 for any time interval n. Because the end goal of our simulation is to calculate the cumulative downtime throughout the entire year, we define ( )t ndownD as the contribution to downtime due to time interval n, as follows:

0

( )( )

( )t nP P n

E n 1down

load in

Acc

AccD =

--

( ) .E n 01tD -

( ) ,E n 0$* (22)

Equation 22 implies that if the accumulated energy at the end of the nth interval is positive, there is no contri-bution to downtime, whereas, a negative accumulated energy implies a downtime of ( )/( ( )),t E n P P n1 load inD - - - which is added to the down counter ( ) .t ndownD^ h To com-pute the downtime throughout the entire year, the con-tributions of every period throughout the year must be summed to obtain

( )

.tt

t n365 24

n

ndownyr down

# #D

D= e o//

(23)

4.2. A Visual Tool for Solar Panel and Wind Turbine ProvisioningTo allow system designers to provision a supercapaci-tor block and a solar panel based on the needs of their applications, we present a visualization-based tool for downtime analysis. As an example, tdown

yr is plotted in Fig. 6 for a system with a load demand of ;P 2 Wload = this plot allows system designers to choose a wind tur-bine size (x-axis) with the rated power in Watts and a supercapacitor block size (y-axis) with a maximum energy storage capacity in terms of Watt hours (Wh). The average downtime is estimated for single source (Fig. 6(a) and Fig. 6(c)) and hybrid configurations (Fig. 6(b) and Fig. 6(d)).

As a demonstration of our visual tool in Fig. 6, we want to calculate the total downtime in a year for an em-bedded system that consumes a constant 2 W to oper-ate, under different power source scenarios:

■ Solar only (Fig. 6(a)): If we assume a 60 W solar panel (P 60 Wsolar

rated = ) and a supercapacitor block of ESC

max = 25 Wh, the intersection of the x-axis and y-axis lands in the contours between 320 and 640 hours. Therefore, our estimated downtime is 480. hours for the entire year (20 days out of 365). Because 25 Wh equals 90000 Joules, this su-percapacitor block would require approximately 8 Maxwell 3000 F individual supercapacitors [50]. To reduce the downtime, a larger supercapacitor block can be used; for example, increasing the supercapacitor block size to 30 Wh (108 kJ) re-duces the downtime to approximately 160 hours

(. 7 days), requiring 10 Maxwell 3000 F superca-pacitors. Alternatively, the same downtime can be achieved by using the same supercapacitor block and a larger solar panel with .P 90 Wsolar

rated =

■ Wind only (Fig. 6(c)): To keep the downtime at ap-proximately 600 hours (25 days) within a year for an embedded system that has the same superca-pacitor block ( ESC

max = 25 Wh), we observe that a wind turbine with the rated power of P 100 Wwind

rated = is needed.

■ Hybrid (P 30 Wwindrated = ) (Fig. 6(b)): In this sce-

nario, we assume a 30 W wind turbine and use a P 30 Wsolar

rated = solar panel. Our goal is to reduce the required supercapacitor block size to achieve the same 480 h downtime, computed in the solar only case, where we use a single P 60solar

rated = W so-lar panel. Figure 6(b) indicates that a 12 Wh super-capacitor block is required to achieve this goal, translating to four Maxwell 3000 F supercapaci-tors. Therefore, we conclude that the usage of the hybrid sources ( W W30 30+ ) rather than solar-on-ly ( W60 ) allows us to reduce the required superca-pacitor block size to half, owing to the complemen-tary nature of the hybrid sources.

■ Hybrid (P 30 Wsolarrated = ) (Fig. 6(d)): In this scenar-

io, we keep the supercapacitor block the same and use solar and wind sources with the same 30 W power rating. We observe from Fig. 6(d) that our downtime reduces to . 40 hours ( 2# days) in a year. Therefore, we conclude that using comple-mentary hybrid power sources without a reduc-tion in supercapacitor size provides a drastically reduced downtime.

5. Energy Harvesting SystemsAlmost all energy harvesting systems leverage the same architecture, shown in Fig. 7. We first investigate this architecture in Section 5.1 and then discuss how vari-ous hybrid harvesters can be implemented atop of this architecture in Section 5.2.

5.1. Harvesting ArchitectureThe general architecture of a conventional energy harvester is depicted in Fig. 7, which consists of four components: (i) power sources, (ii) pre-conditioner, (iii) harvester, and (iv) energy buffer. Some of these components may not exist in a set of harvesters; for example, pre-conditioning is not necessary for most of the solar harvesters. We now examine the details of each component.

Power sources: this component comprises the har-vestable source and a transducer to convert it to electri-cal power. Our proposed systems in this paper include


solar panels and wind turbines as power sources. A multitude of other power sources and transducers are investigated in the literature [44], [86], [87]. Throughout the rest of this paper, we will use the term single-source to refer to solar-only or wind-only; alternatively, we use multi-source to refer to systems with more than one pow-er source. Note that solar/solar is considered a hybrid source, because, for example, two solar panels oriented at different angles can provide power in a complemen-tary manner.

Pre-conditioner: pre-conditioning serves three main purposes: (i) a voltage rectifier turns the AC power in-puts to DC, which is what the harvesting component re-quires; this is necessary for power sources that generate

an AC electrical power, such as a wind turbine as shown in Section 3, (ii) a multiplexer allows the selection of one of the power inputs, (iii) a voltage limiter component prevents the harvesting component from being dam-aged due to over-voltage. For n-power harvesters [30] that generate voltages in the mV range, a voltage limiter is generally not necessary, for example, to power the IoT devices (e.g., Wireless Body Area Networks) found in Medical Cyber Physical Systems [88]–[90]. However, for the wind (and some solar) power sources described in this paper, the generated voltage can exceed 100 V requiring a Pre-conditioning component to incorporate the power sources in typical harvester circuits that use 25–30 V MOSFETs and diodes.

45

40

35

30

25

20

15

100 10 20 30 40 50 60 70 80 90 100

P rated (W)solar

Em

ax (

W. h

)45

40

35

30

25

20

15

100 10 20 30 40 50 60 70 80 90 100

P rated (W)solar

Em

ax (

W. h

)

45

40

35

30

25

20

15

100 10 20 30 40 50 60 70 80 90 100

P rated (W)wind

Em

ax (

W. h

)

45

40

35

30

25

20

15

100 10 20 30 40 50 60 70 80 90 100

P rated (W)wind

Em

ax (

W. h)

1,280

640

320

160 2010

1,280

640

320160

8040

38419224

768384192964824129

6 612 1224 244848 4896

96 96192192 192

384

768

384 384

768

768

1,536

1,536

38419224

2412

64896192384

768 6612

12242448

48 489696 96192

192 192384

768

384 384

(a) (b)

(c) (d)

Figure 6. average downtime tdownyr^ h contours in hours/year for a system requiring a Pload of 2 W for single source (fig. 6(a) and

fig. 6(c)) and hybrid configuration (fig. 6(b) and fig. 6(d)). the average downtime is estimated for a location in rochester, ny region based on the recorded solar irradiation and wind speed data during the years 2009–2010 [60], [83]. In single source sce-narios, tdown

yr decreases as the size of the power source Prated^ h and maximum energy storage EmaxSC^ h increase. however, adding a

second source (wind or solar) with a reasonable size, reduces tdownyr by an order of magnitude without the need to significantly in-

crease the energy buffer size. (a) single source solar harvesting .P 0 Wwindrated =^ h (b) hybrid harvesting with .P 30 Wwind

rated = (c) single source wind harvesting P 0 Wsolar

rated =^ h and (d) hybrid harvesting with .P 30 Wsolarrated =


Harvester: this component is responsible for turn-ing the input power into buffered energy (i.e., harvest-ing). Although the harvester can be implemented using basic hardware elements that do not require software to operate (e.g., OPAMPs, comparators, RLC), for great-er efficiency, intelligent configurations are typically controlled by firmware running on a microcontroller or a DSP. The harvesting component incorporates not only the energy harvesting hardware (e.g., inductors, capacitors, switching MOSFETs), but also the firmware to execute the algorithms, such as the MPPT algorithm introduced in Section 2.2; having software-based intel-ligent harvesters, including the very one we introduce in this paper, provides additional functionality such as over-voltage protection, voltage regulation, and Blue-tooth communication with the embedded device that is being powered.

Energy buffer: buffering the harvested energy is crucial to system’s sustainability, as discussed in Sec-tion 1. A plethora of energy buffering solutions are stud-ied in the literature [54]; batteries and supercapacitors are commonly used, either individually or in combina-tion. One of the responsibilities of the firmware in the harvester is to implement over-voltage protection for the energy buffer; for example, a series connection of 8 Maxwell 3000 F supercapacitors is employed in [50], which has a voltage limit of . 20–22 V, where the vari-ability is determined based on the manufacturing toler-ances of the supercapacitors. Therefore, a software-set over-voltage limit must be imposed in the harvester to

prevent damage to the supercapacitors, which, in some situations, may also be potentially hazardous. Note that this is analogous to the over-voltage limits that must be imposed for rechargeable batteries [91].

Embedded System: the harvested energy is ultimate-ly used to power an autonomously operating embedded system. Examples of such systems include medium-pow-er sensors such as cameras for embedded and distrib-uted computer vision applications. Additionally, lower power sensors can be paired with on-node processing devices such as Digital Signal Processors (DSPs), micro-controllers, and tablets to provide on-site feature extrac-tion [92] and noise reduction capabilities [93], which are often used to overcome data transfer limitations. Addi-tionally, network gateways that are used in low-power WSNs often consume orders of magnitude more power than individual nodes, and these may form the load pow-ered by a medium-power harvester.

5.2. Taxonomy for Energy Harvesting SystemsA large family of harvesting systems that we detail in the following sections are capable of harvesting energy from one or more power sources. To categorize these systems in a concise manner, we introduce a taxonomy that distinguishes each harvesting system based on their power sources. Systems that solely depend on a solar power input are denoted using an S symbol, which means solar-only power input. Similarly, we use W to re-fer to wind energy harvesting systems that depend on a wind-only power input. To extend this notation to include

Wind Turbine Voltage Rectifier

Multiplexer

Voltage Limiter

Pre-Conditioning Harvesting Load

BufferingSolar Panel

Power Sources

Sensing COM OV Protection

OC Protection

Regulator Wakeup Energy Transfer

I

MPPT

V

SW

Battery

Supercap

HW

SEPIC

Figure 7. a typical energy harvesting system is composed of five components: (i) Power Sources include transducers that out-put the harnessed power as electrical power; (ii) Pre-conditioner performs necessary rectification and over-voltage limiting; (iii) Harvester contains the energy harvesting hardware (hW) and firmware to execute harvesting algorithms and provide features such as over-voltage (oV) and over-current (oc) protection; (iv) Energy Buffer buffers the harvested energy using batteries or supercapacitors. (v) Load represents a sensing node or an actuator, with processing and communication capabilities.


systems with more than one power input, we use the no-tations SW (solar and wind), SS (double solar), and WW (double wind), which imply two power sources, and SSW (double solar and wind), SWW (solar and double wind), and SSS (triple solar), which imply three power sources.

In a practical implementation, WW represents a har-vesting system that includes two independent wind tur-bines. Although a single larger wind turbine can also be used in a W configuration, having two wind turbines that are physically placed at two locations with differ-ent daily/hourly wind patterns can allow each turbine to supply a power input with complementary character-istics; therefore, although each individual power source is of the same type, their combination can be thought as being a hybrid power source. Similarly, the WWW configuration can allow a power input from three com-plementary wind power sources. Furthermore, the WW and WWW categories can be used to incorporate a fault tolerance advantage into the system, when compared to W configuration; for example, if one of the wind tur-bines breaks and stops providing a power input in a WWW configuration, the other wind turbines can still be used to generate some power, as a WW. The SS and SSS configurations can be used to gain similar advantag-es; solar panels placed at different angles can provide complementary solar power, while using multiple power panels can also make the system fault tolerant against solar panels failures. The SSW and SWW configurations can further improve the fault tolerance and hybrid input advantages; for example, the SSW configuration not only allows for two solar panels placed at different angles but also takes advantage of sources with inherently comple-mentary characteristics.

We introduce the open source UR-SolarCap [34] so-lar-only (S) harvester in Section 6 and the hardware/firmware modifications to operate as a wind only (W) harvester. In Section 7 and Section 8, we will introduce designs to allow not only multiple power sources (multi-source) but also harvesting systems that use multiple UR-SolarCap boards (multi-board). In Section 9, we will introduce the multi-source single-board design variant.

6. Single-Source Harvesting SystemsAll of the designs that we present throughout the rest of this paper use the UR-SolarCap system [34] as their building block; UR-SolarCap is an open source solar-only (S) harvester, which is originally designed to buf-fer its harvested energy into a supercapacitor block and provide a fixed 5 V voltage output for a target medium-power (1-10 W) embedded device. In [34], the UR-So-larCap is presented as a system comprising (i) a solar panel input, (ii) the UR-SolarCap board itself, which is a 20 cm × 12 cm Printed Circuit Board (PCB) that has

soldered RLC components, a microcontroller and other ICs (hardware), MPPT and other software programmed in the microcontroller flash ROM (firmware), and (iii) a supercapacitor block, as shown in Fig. 8, with the super-capacitor block abbreviated as “SCap” and UR-SolarCap board is abbreviated as “URSC.” This is consistent with Fig. 7, with the exception of the Pre-conditioning com-ponent that is necessary for wind power sources as ex-plained in Section 6.2. Table 6 tabulates the notations that we use to describe the hybrid energy harvesters.

The designs that we introduce in this section and the following sections expand on this concept as follows; (i) they modify the power input to cover a range of hybrid options, including ones introduced in Section 5.2, (ii) they use one or more UR-SolarCap boards in their design for single-board and multi-board options, and (iii) they use the supercapacitor block in exactly the same way. Al-though UR-SolarCap is technically a “system,” we use the term “board” during our descriptions to refer to it in our modularized designs. Therefore, our reference to “multi-board” implies “using multiple UR-SolarCap boards.”

+

–

VSolar

UR-SolarCapBoard

SupercapacitorBlock

SCapURSC

Figure 8. ur-solarcap [34] is a solar-only (S) harvester (de-picted as URSC hereafter), with a single solar input; its buff-ered energy is stored into a supercapacitor block, designated as SCap throughout the rest of the figures in this paper.

Table 6. Notations used to describe the hybrid energy harvesters (Sections 6 and 8), with references to which equations/sections they appear in.

Notation Equations (Sections) Description

Pgenerator 24 (6.3) Output power of a generator

Pharvested 24 (6.3) Power absorbed by a harvester

Psource 24 (6.3) Available power at power generator

Pceil 24 (6.3) 25 (8.2)

Max. power can be absorbed

Pcooperative 25 (8.2) Power harvested in Cooperative mode.

Pregular 25 (8.2) Power harvested in Regular mode.


6.1. Single-Board Solar-Only (S) Energy HarvestingUR-SolarCap can be thought of as an S harvester, accord-ing to the taxonomy introduced in Section 5.2. Modularity of UR-SolarCap allows us to use any type of conventional solar panel at its input, provided that the power and volt-age specifications are not exceeded [34]. As its primary functionality, UR-SolarCap supervises the entire harvest-ing process and distributes the generated power among the load and various internal circuit devices of the sys-tem; it buffers harvested energy into the energy storage block to help reduce system downtime, as discussed in Section 4. Its firmware component executes the fractional VOC method [94].

The MPPT algorithm is executed by the 8-bit PIC16F1783 microcontroller [95], which is a part of UR-SolarCap. The fractional-VOC MPPT algorithm requires the continuous computation of .V V0 82 OCMPP #= [34]. Using a voltage sensor and an integrated ADC, the microcontroller mea-sures the VOC and keeps the solar panel voltage Vsolar^ h close to ;VMPP because the harvester portion of UR-Solar-Cap is built by using a SEPIC DC-DC converter [96], which facilitates energy transfer from the solar panel into the supercapacitor buffer via a single duty cycle parameter (D), a higher D value increases the energy transfer rate, thereby lowering the input voltage Vsolar^ h. Therefore,

when solar irradiation is higher, VMPP consequently be-comes higher, forcing the SEPIC converter to transfer en-ergy at a higher rate.

In addition to its energy harvesting functionality, UR-SolarCap also provides overcharge protection, voltage regulation for internal components, a Bluetooth and RS232 communication infrastructure, and a smooth auto wake-up feature. Its communication infrastructure allows UR-SolarCap to transmit its internal system status—in-cluding the amount of energy available in the supercapaci-tors—to the host embedded device that is being powered from it. Its auto wake-up feature allows it to shut down for an indefinite amount of time (e.g., during multiple consec-utive days with no solar power input) and resume a fully functional status at the end of this dark period. The imple-mentation and architecture of UR-SolarCap are discussed extensively in [34].

6.2. Single-Board Wind-Only (W) Power HarvestingIn this section, we describe how a wind-only energy har-vesting system (W ) can be designed by adapting the solar-only (S ) UR-SolarCap board; our design goal is to retain the versatility of UR-SolarCap by incorporating the auto wake-up, over voltage and over power protection, Bluetooth and RS232 communication, and a regulated 5 V voltage features. Furthermore, we introduce a new ver-sion of the PIC firmware that executes an MPPT algorithm, which is suitable for wind power sources. Implementing a W system proves to be more complicated than simply connecting an S harvester to a wind turbine, because the characteristics of wind and solar power sources are fun-damentally different. Nonetheless, it is possible to reuse the UR-SolarCap hardware and software as the basis for a W harvester design.

The high-level architecture of our proposed W con-figuration is shown in Fig. 9. Compared to the S harvest-er, the proposed W harvester includes three additional hardware components: (i) a wind turbine to convert the mechanical wind power to electrical power, (ii) a recti-fier to convert the three-phase AC output of the turbine to DC, and (iii) a 25 V voltage limiter to ensure that the UR-SolarCap voltage specifications are not exceeded. Figure 10 depicts the circuit schematic of the rectifier and the voltage limiter, which we envision as a daugh-ter board that interfaces UR-SolarCap; this modular de-sign eliminates the need for extensive modifications to UR-SolarCap to turn it into a W harvester. The voltage limiter is necessary because the voltage output of the wind turbine can reach as high as 50 V, posing a damage threat to UR-SolarCap. This contrasts with n-power sys-tems, such as Medical Cyber Physical Systems, which incorporate IoT components that produce voltages far below 1 V [97]–[100].

ac dc dc+ +–

–

Vturbine URSC

SCap

ac dc

L1L2L3

RectifierVoltageLimiter

Figure 9. In the W configuration, the ac output of the wind turbine is first rectified and voltage-limited to comply with ur-solarcap voltage specifications. additionally, a new firm-ware effectively tracks the mPP.

Rectifier Voltage Limiter

D1

Vin1 Vin2 Vin3

D2

D3

D4

D5

D6

C

R

T1

Dz22 V

vout+

vout–

Figure 10. the output of the voltage rectifier may reach .50 V2 this circuit limits it to . 25 V, which is required by

ur-solarcap. When Vin reaches 24–25 V, the zener diode Dz turns on ,T1 thereby shunting the turbine and clipping its voltage. terminal Vin is connected to the wind turbine. terminal Vout is connected to the ur-solarcap main board supply input.


Since the P-V characteristics of solar and wind are dif-ferent, an alternative MPPT algorithm is needed for the W configuration. A variety of such algorithms has been studied in the literature [101]–[104]. Our proposed W architecture employs the well-known Hill Climbing (HC) method [105]. The implementation of the HC algorithm is simple, and does not require wind or rotational speed sensors. However, its capability to track the MPP in rap-idly changing wind conditions is limited [106]. Because the S and W designs use different MPPT algorithms, a new firmware design is necessary for UR-SolarCap to al-low it to work as a W harvester.

The power output of a prototype implementation of a UR-SolarCap-based W harvester—that uses a Higoo™ 50 W turbine as its wind power generator—is shown in Fig. 11 for a five-minute interval; the data is logged through the volt-age and current measurement capabilities of UR-SolarCap and transmitted to a PC through its RS-232 communication module. The transmitted data is then saved as a CSV and plotted using GNU Octave [107]. Figure 11 demonstrates an interesting characteristic of wind power supply; while solar supply exhibits long periods of steady power, wind supply exhibits large peaks in short intervals.

6.3. Single-Board Single-Source CharacteristicsThe main advantages of the S and W configurations are their simplicity and low cost, especially in environments with high solar and wind power availability. However, single-board single-source harvesting setups are sus-ceptible to two fundamental drawbacks: (i) the original UR-SolarCap design imposes a power limit Pceil^ h to avoid damage to its internal components. Therefore, even if the available power at the power generator Psource^ h is higher than ,Pceil the harvested power P harvested^ h is lim-ited to Pceil by the harvester; as formulated below:

( , ),minP P Pharvested source ceil= (24)

which can lead to substantial waste when .P Psource ceil&

Moreover, (ii) as mentioned previously, reliance on a single energy source restricts the applicability of a field system to locations where solar or wind energy are con-tinuously available. In the following sections, we intro-duce harvesting configurations that can utilize hybrid power sources.

7. Hybrid Harvesting SystemsIn this section, we introduce three distinct hybrid (multi-source) harvesting system architectures that can harvest and buffer a combination of solar and wind power sources.

(i) Independent Multi-Board harvesting: This archi-tecture incorporates multiple non-communicating

harvesting boards, each responsible for harvesting a single power source. We expand our taxonomy introduced in Section 5.2 to cover the configura-tions in this category; for example, SWW refers to an independent multi-board configuration that in-cludes one solar panel and two wind turbines. We provide an overview of this category in Section 7.1.

(ii) Cooperative Multi-Board harvesting: This archi-tecture incorporates multiple communicating har -vesting boards, where each board not only har-vests its own power source, but also communicates with the other boards for improved efficiency, fault tolerance, and extended power range. S S2 2 is an ex-ample of this category, where two harvesting boards communicate to jointly harvest two solar sources. We provide a high-level overview of cooperative multi-board harvesting in Section 7.2; furthermore, a detailed operational analysis of this category is pro-vided in Section 8.

(iii) Time-Multiplexed Single-Board harvesting: This architecture employs a single harvesting board, which can receive power from two or more sources through an analog multiplexer; each mul-tiplexed input is harvested only a fraction of the time, hence our naming time-multiplexed. An exam-ple notation “S W W ” refers to an architecture in this category that receives one solar and two wind power sources as input. We provide an overview of this architecture in Section 7.3 and a detailed op-erational analysis in Section 9.

7.1. Independent Multi-Board HarvestingIn this architecture, the outputs of multiple single-source harvesters (i.e., S or W configurations) are connected

30

25

20

Inpu

t Pow

er (

W)

15

10

5

01 2 3 4 5 6

Time (min)7 8 9 10

InstantaneousAverage

Figure 11. Wind energy harvesting using a firmware-modified ur-solarcap board. the short peaks in the wind power sup-ply are in contrast with the relatively steady solar power input.


to the same supercapacitor block. While each harvest-er receives its input power from a different source, its harvested energy is stored in a shared supercapacitor block. Representative configurations of this family are shown in Fig. 12; for example, the SWW configuration in Fig. 12(e) utilizes three separate UR-SolarCap boards to harvest one solar and two wind power sources indepen-dently and the harvested energy is buffered in a single supercapacitor block to power an embedded device.

In the SS and SSS configurations, two or three sepa-rate solar panels provide power into the system. Al-though these two configurations use the same type of power source, they can still be considered as hybrid power sources, because installing them at different lo-cations and orientations can produce a complementary power input; for example, for two solar panels facing the sun at different angles, the power can increase at one so-lar panel when it decreases at the other [108]. A similar argument for the WW and WWW configurations can also be made; for example, in mountainous areas, multiple wind turbines—placed at different physical locations— can provide complementary characteristics.

In comparison to other hybrid architectures, simplic-ity is the main advantage of independent multi-board harvesting. Because of the independence of the harvest-ers, this configuration is highly scalable; adding a new power input (e.g., going from WW to SWW) is just the matter of adding another UR-SolarCap and connecting its input to the new power source (S in this example)

and its output to the shared supercapacitor block. Fur-thermore, this scalability can be used to provide fault tolerance to the system; if one of the harvesters fails, the others continue harvesting energy, thereby avoiding a total system shutdown.

7.2 Cooperative Multi-Board HarvestingThe cooperative multi-board harvesting architecture consists of two single-source (S or W) harvesters that communicate through a link to coordinate their opera-tion. Figure 13 depicts the configurations in this catego-ry; the communication link—used in cooperation during harvesting—is denoted as “COM” in Fig. 13. Each UR-So-larCap board has an upper limit in the amount of power it can harvest, denoted as ;Pceil the internal firmware over-power protection prevents a UR-SolarCap board from harvesting a higher amount of power to avoid cir-cuit damage. The cooperation mechanism introduced in this category allows two boards to harvest 2× the amount of power of a single board. To achieve this, an analog multiplexer is placed between each power source and UR-SolarCap board, allowing that power source to be channeled into one of the UR-SolarCap boards; a se-lection logic, determined as a function of the coopera-tion mechanism, enables this architecture to be used in one of two operational modes: Regular and Cooperative.

In Regular mode, which is the system default, the system operates as an independent multi-source configuration, in the way described in Section 7.1,

URSC

URSC URSC URSC

URSC

URSCURSC

URSC

URSCURSC

URSC

URSC

URSC

URSC

URSC

SCap

SCap

SCap SCap

SCapSCap

(a) SS (b) SW (c) WW

(d) SSW (e) SWW (f) WWW

Figure 12. a high-level overview of independent multi-board harvesting configurations that we introduce in section 7.1. the ur-solarcap board is denoted as “ursc,” which receives its input from one or more solar/wind power sources and buffers its harvested energy into a supercapacitor block, which is denoted as “scap.”


by selecting the multiplexers to channel each power input to its corresponding harvester. In case one of the boards has a higher amount of power available at its input than ,Pceil it communicates this status to the second board; the second board then joins efforts to harvest the same input cooperatively, by selecting its multiplexer to channel that power input to itself. This mode, where a single power input is channeled into two harvesters, is called the Cooperative mode, dur-ing which both harvesters use the COM link to adjust their duty cycles to harvest an equal portion of the available power input. We provide the implementation details of this mode in Section 8.

S2 and W2 options include a single power source, which is jointly harvested by two boards, whenever the system operates in Cooperative mode. In Regular mode, however, one board harvests its power source and the other one remains idle. Rather than being a hybrid sys-tem, S2 and W2 are used in cases where solar panels and wind turbines provide power inputs that are too high for a single board to handle; in these cases, the system always operates in Cooperative mode, thereby avoiding a waste of resources. In Regular mode, ,S S W W2 2 2 2 , and S W2 2 configurations operate as SS, WW, and SW har-vesters, respectively. In Cooperative mode, the power

source that provides a lower amount of power is discon-nected form the system using the multiplexer, while the other source is jointly harvested by both boards by dis-tributing the input power to them evenly.

Among all of our proposed hybrid architectures, co-operative multi-board is the only one that addresses the power limitation Pceil^ h of the S and W configurations. Assuming that the both boards in this configuration are identical, the harvesting power limit of the entire system reaches .P2 ceil# Moreover, similar to independent multi-board harvesting configurations, described in Section 7.1, cooperative multi-board harvesters offer fault-tolerance through hardware redundancy. However, as we discuss in Section 8, Cooperative mode adds to the system com-plexity in terms of both the hardware and firmware. Al-though we only describe and report on experiments for this category with two harvesters, an expansion to 3$ boards is possible.

7.3. Time-multiplexed Single-Board HarvestingAlthough the hardware redundancy and the increased power limit, offered by the previous two categories, is preferable in applications where cost is not of primary concern, we propose the time-multiplexed single-board harvesting architecture for cost sensitive applications.

Reservedfor

Scalability

Reservedfor

Scalability

URSC

URSC

URSC URSC

URSCURSC

URSC

URSC

URSC

URSC

CO

M

CO

M

CO

M

CO

M

CO

M

SCap

SCap SCap

SCap SCap

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

SEL[1:0]

MU

X 2

:1M

UX

2:1

MU

X 2

:1M

UX

2:1

MU

X 2

:1M

UX

2:1

MU

X 2

:1M

UX

2:1

MU

X 2

:1M

UX

2:1

(a) S2 (b) W2 (c) S2S2

(d) W2W2 (e) S2W2

Figure 13. In the cooperative multi-board harvesting architecture (introduced in section 7.2), two ur-solarcap boards (ursc) can jointly harvest a single power source to generate a higher amount of power; the S2 and W2 configurations are the special cases of this category, geared towards harvesting more energy rather than hybrid power input. alternatively, the other configu-rations , ,(S S W W2 2 2 2 and )S W2 2 add hybrid power input functionality. the “com” link represents the communication channel between the two ur-solarcap boards used for cooperative harvesting.


Harvesters in this category utilize a single harvesting board to simultaneously harvest multiple power sourc-es. As depicted in Fig. 14, every configuration in this category consists of (i) two or more power sources, (ii) an analog multiplexing mechanism, (iii) a single harvest-ing board, and (iv) a supercapacitor-based energy buf-fer. The solar and wind power inputs are connected to the harvesting board through the analog multiplexer, which is controlled by the firmware; the firmware cycles through the available power inputs in a round-robin fash-ion by connecting a single input at any point in time and disconnecting the others. While a power source is dis-connected, its generated energy is buffered in a small ca-pacitor. When connected, the harvester treats the small capacitor as the power source and harvests from it dur-ing the allocated time slot for that source; in this way, the harvesting process is time-multiplexed. A detailed analysis of the time-multiplexed single-board harvesting category is presented in Section 9.

The primary advantage of this category is its re-duced costs and potentially reduced system size due to the elimination of multiple boards. Furthermore, ex-pansion to multiple power sources using this category is fairly straightforward. However, this category suffers from a potentially increased hardware failure rate, due

to the harvesting board being a single point of failure in the system.

7.4. Comparison of Harvesting ArchitecturesA comparative summary of all harvesting architec-tures introduced in this section is tabulated in Table 7. Overall, due to its simplicity and fault tolerance, in-dependent multi-board harvesting is suitable for fast deployment of mission-critical systems. Cooperative multi-board harvesting is recommended, if the nomi-nal power of an existing power transducer such as a solar panel or wind turbine exceeds the harvesting capability of the harvesting board. This is a common design practice, as system designers are occasionally compelled to tradeoff cost for system availability to meet the worst-case scenario requirements. In normal scenarios, however, cooperative multi-board harvest-ing can harvest the surplus portion of energy to sup-port auxiliary functionality. Time-multiplexed single-board harvesting is suitable for large-scale systems, in which fault tolerance is not a requirement. The expandability of this architecture also adds to sys-tem flexibility by allowing each harvesting node to be modified based on the power requirements driven by desired new functionality.

MU

X 2

:1M

UX

3:1

MU

X 3

:1

MU

X 3

:1

MU

X 2

:1

MU

X 2

:1URSC

URSC URSC URSC

URSC URSC

SCap

SCap SCap SCap

SCap SCap

SEL[1:0]

SEL[1:0] SEL[1:0] SEL[1:0]

SEL[1:0]SEL[1:0]

(a) SS (b) WW (c) SW

(d) SSS (e) WWW (f) SSW

Figure 14. some example configurations of the time-multiplexed single-board harvesting system (introduced in section 7.3); in all of these configurations, a single harvesting board concurrently harvests multiple power sources. When a power source is not connected to the harvesting board, its generated power is buffered in intermediary capacitors, inserted between the power source and the analog multiplexer (which are termed input buffer capacitors). In this figure, ursc box designates the harvesting board and scap represents the supercapacitor-based energy storage block.


8. Hardware/Firmware Implementation: Cooperative Multi-Board Harvesting

In this section, we detail the implementation of cooper-ative multi-board harvesters introduced in Section 7.2, which are composed of five components: (i) two single-board single-source harvesting boards, (ii) a commu-nication channel to allow the two boards to communi-cate, (iii) a shared supercapacitor-based energy buffer, (iv) one or more power inputs that are cooperatively harvested by these boards, and (v) an analog multi-plexer per board to channel the power sources into these boards in a variety of configurations, determined by the harvesting mode. Whenever we are referring to configurations with multiple boards, we denote them as board1, board2, etc. With respect to hardware, the two harvesting boards utilized in cooperative multi-board harvesters are identical to the S and W con-figurations. However, their software operation differs based on the mode that they are currently operating in (Regular and Cooperative, as discussed in Section 7.2).

In this section, we explain the implementation details of the hardware and software components of coopera-tive multi-board harvesters.

8.1. Analog Input Multiplexing CircuitryThe analog multiplexer, connected to the input of each harvesting board, allows the channeling of one or two power sources into the harvester. In the S2 and W2 config-urations (Figs. 13(a) and 13(b)), although a simple switch will suffice, they are indicated by 2:1 MUXes, where one input is not connected to any power source; instead, it is reserved for connecting a power source in the future. This depiction generalizes this category; an additional power source can be readily connected to the available reserved MUX input for scalability. Configurations with two power sources such as S W2 2 or W W2 2 (Fig. 13(e) and Fig. 13(d)), also use 2:1 MUXes.

An implementation of analog multiplexing functional-ity using N channel and P channel MOSFETs is depicted in Fig. 15, where the inputs SRC1 and SRC2 are connected

Table 7. Example configurations of hybrid harvesters proposed in the paper, along with the section each configuration is described in and its associated figure. Each ( • ) represents a single power source included in the configuration; for example, SS harvester incorporates two separate solar power sources. Board Count represents the number of UR-SolarCap harvesting boards in each configuration.

Type Sec. Notation SourcesBoard Count Fig.

S W

Independent Multiple Board 7.1 SS •• 2 12(a)

SW • • 2 12(b)

WW •• 2 12(c)

SSW •• • 3 12(d)

SWW • •• 3 12(e)

WWW ••• 3 12(f)

Cooperative Multiple Board 7.2 S2 • 2 13(a)

W2 • 2 13(b)

S S2 2 •• 2 13(c)

W W2 2 •• 2 13(d)

S W2 2 • • 2 13(e)

Time Multiplexed Single Board 7.3 SS •• 1 14(a)

W W •• 1 14(b)

S W • • 1 14(c)

S S S ••• 1 14(d)

W W W ••• 1 14(e)

S WS •• • 1 14(f)


to the power sources. Their outputs, OUT1 and OUT2, are connected to the harvesting boards. The channeling of the power sources into the SRC inputs of each multi-plexer is achieved by a four-bit selection signal as shown

in Table 8. Each power source can be channeled into board1, board2, or left disconnected (Hi-Z). For example, in Regular mode, the 0101 selection vector establishes the SRC OUT1 1" and SRC OUT2 2" connections, there-by allowing board1 to harvest the first power source and board2 the second. Alternatively, when the power level at SRC2 becomes much higher than SRC1, the firmware enters the Cooperative mode and selects the 0110 vector, establishing the SRC OUT2 1" and SRC OUT2 2" con-nections; in this case, both board1 and board2 cooper-ate towards harvesting an equal portion of the available power. The firmware is designed to never select one of the “Not Allowed” conditions, because they may cause hardware malfunction.

The multiplexing mechanism works as follows: Apply-ing a “0” to any one of the SEL inputs turns off the cor-responding N channel MOSFET; in a short time period after this, the P channel MOSFET—acting as a switch—turns off. For example, applying SEL1=0 turns off T1 and causes the charge in the CGS of the P channel MOSFET—depicted as S1 in Fig. 15—to drain and its VGS go down to ,V V0GS = thereby turning it off. The amount of time that it takes between setting the SEL1=0 to the time when S1 is totally turned off is determined by the RC constant of R1 and the input capacitance of .S1

To turn S1 back on, SEL=1 must be applied, which will re-charge the CGS of S1 in an RC constant determined by the input capacitance of ,S1 as well as the ,R1 R2 pair. The voltage drop across these resistors results in

,V V0GS 1 which turns on the P channel MOSFET and sets its output voltage approximately equal to its source volt-age. Depending on the voltage of the connected source, the voltage of the output might exceed the disconnected source voltage. Even if the P channel MOSFET is off, some

SRC1

SEL1

SRC2 SRC2 SRC1

SEL2 SEL3 SEL4

R1

R2

R3

R4

R5

R6

R7

R8

T2T3 T4T1

S1 S2 S3 S4D1 D2 D3 D4

OUT1 OUT2

(a) (b)

Figure 15. analog multiplexers used in the cooperative multi-board hybrid harvesting architecture. sel terminals are set or re-set by the microcontroller according to table 8. resistors R R1 8f are used to properly bias the P channel mosfets ( ).S S1 4f diodes D D1 4f are used to prevent drain to source current in these mosfets. the relationship between the output and se-lect signals is tabulated in table 8. In our implementations, we used 5.6 kR R R R .1 3 5 7 X= = = = ,2.5R R R R k2 4 6 8 X= = = =

2 7002,T T T T N1 2 3 4= = = = 9540 ,S S S S IRF N1 2 3 4= = = = and 560.D D D D SB1 2 3 4= = = =

Table 8. The relation between the output and select signals of the multiplexers in the cooperative multi-board harvesting architecture. “SELECT” represents the four control signals denoted by SEL1, SEL2, SEL3, and SEL4 in Fig. 15. “Not Allowed” combinations should be avoided, because they are either unsupported by the hardware or are undefined in the firmware. The SRC1 and SRC2 inputs are the primary sources of harvesting Board1 and harvesting Board2. Combination 1010 is Not Allowed, because in cooperative multi-board harvesting, at least one of the harvesting boards must harvest its primary power source.

SELECT OUT1 OUT2 Mode

0000 HI–Z HI–Z OFF

0001 SRC1 HI–Z Regular

0010 SRC2 HI–Z Not Allowed

0011 X HI–Z Not Allowed

0100 HI–Z SRC2 Regular

0101 SRC1 SRC2 Regular

0110 SRC2 SRC2 Cooperative

0111 X SRC2 Not Allowed

1000 HI–Z SRC1 Not Allowed

1001 SRC1 SRC1 Cooperative

1010 SRC2 SRC1 Not Allowed

1011 X SRC1 Not Allowed

11XX X X Not Allowed


current might still flow from the output to the source through the transistor’s body diode. Diodes D1 and D2 are added to the design to prevent the reverse current from output to the source ports in these situations.

8.2. Cooperation MechanismWe use the term cooperation mechanism to refer to the platform that allows the boards to communicate and coordinate their harvesting operation. This mechanism involves two layers: (a) a hardware-implemented physi-cal layer and (b) a firmware-controlled layer. The physi-cal layer is composed of two serial data communication wires, RX and TX. As there are only two harvesting boards in cooperative multi-board configurations, no advanced networking services such as access manage-ment and node addressing is required. However, if the architecture needs to be adapted to include more than two harvesting boards, more sophisticated communica-tion protocols (e.g., SPI and CAN) can be utilized. The firmware layer includes a simple control protocol for the two boards to synchronize their harvesting; this protocol manages the physical layer and allows the two boards to determine when to initiate/terminate/accept/reject the Cooperative mode.

In cooperative multi-board harvesting, the firmware of each harvester starts its operation in Regular mode and is responsible for determining when to switch to Cooperative mode; when the power level at a source in-creases to beyond ,Pceil each board requests the other to join its efforts to harvest this source cooperatively, through a four-stage handshake protocol: (i) the initia-tor (board1) first sends a request by setting its TX chan-nel, (ii) board2 suspends its MPPT if its input power is below a threshold and switches its power source by placing the proper MUX select signals; it provides an acknowledgment to board1 by setting its TX pin (which is connected to RX pin of board1), (iii) once its RX pin is set, board1 enters the Cooperative mode and clears its TX pin (which is connected to RX pin of board2), and (iv) finally, board2 begins the Cooperative mode as soon as its RX pin is cleared. To eliminate ambiguity, we refer to board1 as master and board2 as slave. When both boards are in Cooperative mode, the master determines the right value of the duty cycle and shares it with the slave through the communication channel. This approach guarantees that the harvesting load on both boards is equal because their duty cycles are identical.

Although simple, this suggested cooperation mecha-nism may fail to increase the overall amount of harvest-ed power in multi-source cooperative configurations

, , .S S S W W W2 2 2 2 2 2^ h This limitation is shown by Eq. 25, which compares the range of instantaneous delivered power in cooperative configurations operating in Coop-

erative Pcooperative^ h and Regular Pregular^ h modes. Assum-ing that the cooperation threshold is set to ,Pceil the total amount of harvested power cannot fall below this value when the system is in Cooperative mode. As Eq. 25 implies, it is possible that Pregular exceeds Pcooperative (because they overlap), which means that the system is worse off by sustaining the Cooperative mode in some situations. This is a condition when the master decides to terminate the Cooperative mode; the most straight-forward approach is to exit this mode if the harvested power of each board is lower than half of the .Pceil

,

.P P P

P P2

0 2ceil cooperative ceil

regular Ceil

# #

# # (25)

However, as it is evident from Eq. 25, terminating co-operation even when the input power exceeds half of the Pceil might increase the harvested power in some cases. One solution to this problem is to set higher values for the threshold. This, however, reduces the duration in which the system operates in the Cooperative mode. An-other solution is to continually log the harvested power and use the history of the system to predict the incoming power in near future. Both master and slave can then de-cide to enter or leave cooperation based on their expec-tations of changes in the incoming power. The efficacy of this solution is completely dependent on the accuracy of the power forecasting. Therefore, for sources such as solar power, which are generally stable and predict-able, this approach is compelling. However, predict-ing the wind pattern using the past data proves to be quite challenging.

8.3. Single Source, Multi-Board ConfigurationsThe S2 and W2 setups include only a single power source, which effectively makes them single-source energy har-vesters. In both configurations, the specifications of power sources, particularly the output voltage, should comply with UR-SolarCap original requirements. How-ever, due to the option of cooperative harvesting, the sys-tem can harvest up to twice the amount of UR-SolarCap power limitation .Pceil^ h

8.4. Multiple Source, Multi-Board ConfigurationsIn S S2 2 , the panels are typically set up in different ori-entations to provide a more steady supply of harvest-able energy as the relative location of sun in the sky var-ies. Because the performance of solar panels is highly dependent on the direction of the incident light [108], the enhancement in efficacy provided by double-panel harvesting should not be underestimated. Similarly, two wind turbines placed in different locations can be simultaneously harvested in W W2 2 configuration. It is also possible to harvest two power source types in


S W2 2 setup. The design of multiplexers is identical in , ,S S W W2 2 2 2 and S W2 2 configurations. Two 2:1 analog

MUXes are included to provide a path among all the har-vesting boards and power sources. All the other hard-ware and software components remain unaffected as well. The board that executes the HC MPPT algorithm in its firmware should harvest wind energy as its primary source, while the harvesting board executing fractional open-circuit voltage MPPT method should primarily harvest solar energy. Each board harvests its primary power source in Regular mode and can initiate the Coop-erative mode if the output of its primary source exceeds the pre-defined threshold.

The communication protocol of cooperation mecha-nism limits the system’s scalability to two harvesting boards. However, with additional adaptations in the mul-tiplexer design and communication protocols, this ar-chitecture can be adapted to include more power sourc-es. Particularly, S S2 2 and W W2 2 can be combined into a triple source harvesting, operating as either S S W3 3 3 or S W W3 3 3 systems. In this case, 2:1 multiplexers should be replaced by 3:1 multiplexers. Furthermore, an address-ing protocol should be implemented in the system, mak-ing each board separately addressable. An I C2 protocol can meet this requirement well. Once implemented, triple source cooperative harvesting not only lowers the system’s downtime due to higher number of included power sources, but also increases the system’s power limit to ,P3 ceil# when all boards have an identical Pceil power limit.

9. Hardware/Software Implementation: Time-Multiplexed Single-Board

In this section, we detail the implementation of time-multiplexed single-board harvesters introduced in Sec-tion 7.3. Harvesters of this category have a significant cost advantage because they only require a single UR-SolarCap board to harvest more than one power source. However, due to that single board being a single point of failure, designs in this category suffer a reliability and fault-tolerance disadvantage.

9.1. Time-Multiplexing MechanismAll of the power sources are connected to a single har-vesting board through an n:1 analog MUX, where n is the number of power sources. Voltage limiters and recti-fiers are required if the power source voltage does not comply with UR-SolarCap specifications (Section 6.2). Unlike the cooperative multi-board architecture, the included power sources in a time-multiplexed configu-ration are not connected/disconnected based on their power availability; instead, the harvesting board pe-riodically switches between them—in a round robin

fashion—at a pre-defined frequency .fsw^ h In summary, a time-multiplexed harvester is a polymorphic system that periodically changes between S and W configura-tions; from the perspective of the harvesting board, the system harvests a single power source at every instant of time, similar to S and W setups.

9.2. Analog Input Multiplexing CircuitryAnalog MUXes used in time-multiplexed single-board and cooperative multi-board architectures are identi-cal, although their operation differs as follows. In co-operative multi-board harvesting, the multiplexer op-eration is controlled based on the power availability of each source, while in time-multiplexed single-board systems, the switching occurs periodically regardless of the power level of each source. An n:1 multiplexer is required for a time-multiplexed harvester, harvesting n power sources. For this design, the same 2:1 MUX shown in Fig. 15 can be utilized; on the left side of Fig. 15, a 2:1 MUX is constructed from two separate analog switches that connect SRC1 and SRC2 to the same output (OUT1). This circuit is highly scalable by using the same (N channel–P channel–R–R–D) building block and connect-ing a new input (e.g., SRC3) to the same output (OUT1) to build a 3:1 MUX. Similarly, a fourth repetition of the same building block implements a 4:1 MUX, connecting SRC4 to OUT1, etc. An n:1 MUX requires n repetitions of the same building block.

9.3. Firmware to Implement Time-MultiplexingThe firmware of time-multiplexed single-board harvest-ing systems performs two additional tasks compared with other architectures. First, it controls the multi-plexer to periodically switch among all available power sources. The firmware also executes the proper MPPT algorithm according to the type of each connected pow-er source. The switching process is accomplished using a selection vector, composed of SEL1, SEL2, f signals (Fig. 15). To assure proper operation, the round robin sequencing among available power sources is accom-plished by first disconnecting the currently-in-use power source and then connecting the next. More sophisticat-ed sequencing algorithms than round robin can also be implemented—such as priority-based queuing and inter-rupt-based selection—according to application require-ments. Before disconnecting a power source, the firm-ware saves a snapshot of its MPPT algorithm execution status. Whenever that source is reconnected, its MPPT status is retrieved and the firmware continues the MPPT procedure from that point. Although the MPPT status of a power source changes during its disconnection period, this retrieval of its previous MPPT status substantially improves the firmware efficiency by reducing the MPPT


algorithm delay. This is particularly beneficial for con-figurations that incorporate one or more wind power sources, because wind turbines have a significant MPP convergence delay.

9.4. Harvesting Power LimitThe time-multiplexed harvesting architecture can har-vest multiple power sources simultaneously, nonethe-less the power limit for each source is still .Pceil Further-more, since there is only one harvesting board in this configuration, the total charging power of the superca-pacitor block cannot exceed .Pceil This contrasts with the cooperative multi-board architecture, in which the power limit is P2 ceil# and the independent multi-board configuration—using n harvesting boards—with a pow-er limit of .n Pceil#

9.5. Multiplexing EfficiencyThe non-ideality of the MOSFET switches that are used in the input analog MUX and ripples around VMPP also impact the efficiency of the time-multiplexed configura-tions. MOSFET power dissipation can be categorized into resistive, transient, and switching losses. Resistive power losses are determined by RDSon and VMPP of the currently connected power source. Transient power dissipation oc-curs during ON$OFF and OFF$ON transitions and is proportional to the drive capability of the switching transistor that the input MUX consists of. Switching power losses, on the other hand, are typically negligible due to a low switching frequency (333 Hz) and small number of power transistors (two power transistors for a double-source time-multiplexed harvester). Our anal-ysis and observation confirmed that %95mux .h in the worst-case scenarios.

Apart from the power consumption of the multiplex-er, the fluctuations around VMPP also have a negative impact on the efficiency of time-multiplexed harvesting configurations .swh^ h We can assess the percentage of voltage fluctuations in the worst case scenario [53]. If the size of the capacitors inserted between the source and the switch is 4700 nF, the switching frequency is 333 Hz, and .I 3 48charge

max = A, the voltage of the input buf-fer ripples ±1.11 V. Assuming ,V 13 VMPP = we calculate the ripples to be around 8.5%. This corresponds to an

.0 95sw .h (95%), based on the study in [51].

9.6. System Modularity and ExpandabilityThe modularity of the time-multiplexed design ensures its expandability. Adding new power sources to the time-multiplexed harvesting architecture requires a simple modification in the analog MUX by adding a (NMOS–PMOS–R–R–D) building block to the circuit shown in Fig. 15 (see Section 9.2). Whenever a new power source is

added, the firmware should be notified of its type and its harvesting order in the round-robin scheme; typically, this requires a simple update in the embedded C code, recompiling and reflashing the firmware. Theoretically, any number of power sources can be used in time-mul-tiplexed harvesting architecture. However, having too many sources increases the time interval during which a power source remains disconnected, thereby necessi-tating larger input capacitors and/or a faster switching frequency to prevent large ripples around ,VMPP as ex-plained in Section 9.1. Arbitrary increases in the input buffer capacitance and/or the frequency are, however, also not practical. Furthermore, adding extra power sources without increasing the power limit of UR-Solar-Cap Pceil^ h can saturate the amount of power transferred to the supercapacitor buffer, thereby hindering the nor-mal operation of the system, unless the excess portion of the power can be consumed by the embedded load.

10. ExperimentsIn this section, we provide experimental results for a set of representative architectures in family of harvesters described in Sections 7, 8, and 9. We introduce our ex-perimental test bench setup in Section 10.1. Our experi-ments are conducted at a University of Rochester ECE building through long-term measurement and logging of solar/wind patterns and operation of our harvesters. In Section 10.2, we provide the results for auto wake-up functionality of our W harvester, followed by our evalu-ation of the SW harvester, which is a configuration rep-resentative of independent multi-board harvesting dis-cussed in Section 10.3. To demonstrate the operation of our cooperative multi-board harvesters, we provide experimental results for solar S S2 2^ h and wind W W2 2^ h power sources in Sections 10.4 and 10.5, respectively. Our time-multiplexed single-board harvesting results for the S S configuration are presented in Section 10.6. Table 9 lists configurations on which we have conducted experiments along with the sections where each topol-ogy is introduced and the results are presented.

10.1. Experimental SetupThe experimental setup for hybrid solar/wind energy harvesting with two UR-SolarCap boards is illustrated in Fig. 16. The wind turbine is used as the wind power source, while one or more solar panels are used as the so-lar power source, depending on the power requirement. The multiplexer board provides separate voltage connec-tions for two UR-SolarCap boards, which communicate with each other through dedicated RX/TX wires and con-trol the switches on the multiplexing board through ad-ditional wires. We provide the details of different compo-nents used in our experimental setup below:


Wind turbine: The nominal specifications of the wind turbine are 50 W and 12 Vrms (AC). However, the rectified DC voltage can reach as high as 40 V under no load condi-tion, necessitating the use of a voltage limiter to prevent damage to the UR-SolarCap and multiplexer board.

Solar Panels: A set of solar panels with different rated voltages are shown in Fig. 16. In our experiments, a par-allel configuration of multiple solar panels was used to provide the required solar power for each experiment; for example, two 30 W solar panels were connected in parallel to provide a 60 W rated power. The experiments were carried out between October and December in the Rochester, NY area. Due to relatively low solar irradia-tion during that period, we had to place two solar pan-els in parallel. The solar panels have a maximum open

circuit voltage of 24 V, which is less than the maximum allowed input voltage of UR-SolarCap.

Supercapacitor block: consists of 8 # 3000 F serially connected supercapacitors with a maximum rated volt-age of 2.7 V each. UR-SolarCap firmware features an over-voltage protection by stopping energy harvesting when the supercapacitor block voltage exceeds . .8 2 7 21 6# = V.

Electronic load: is a programmable current sink, which is used to discharge the supercapacitor block when the block is charged to full capacity. The sink cur-rent is usually set to 7 A to discharge supercapacitors in a relatively short period while preventing the wire con-nections from overheating.

Pickit 3: is a programming device, which uploads a new firmware—from a HEX file—into the microcontroller’s

Load Rectifier/Limiter Multiplexing Board SC Block 50 W Wind Turbine

Input Buffer Cap MUXes URSCs 1 × 30 and 2 × 10 W Solar Panels

Figure 16. the experimental hybrid (solar/wind) harvesting setup (fig. 14) used to provide the proof-of-concept results in this paper. a 50 W wind turbine is stabilized by mounting it on a pole that attaches to a wood panel on the bottom; this pole passes through a hole that is drilled on a desk. to avoid vibration, the pole is fixed by multiple screws that attach it to the desk. this setup is capable of providing experimental results for a wide range of the hybrid configurations introduced in this paper.

Table 9. List of the configurations used in our experiments section/figure numbers where the corresponding results are presented.

Harvester Configuration Results

Notation Section Description Figure Section

W 6.2 Wind-only wake up 17 10.2

SW 7.1 Solar and wind Independent 18 10.3

S S2 2 8 Two Solar Panels Cooperative (Independent SEPIC duty cycle computation)

19 10.4

Two Solar Panels Cooperative (Master-Slave) 20

W W2 2 8 Two Wind Turbines Cooperative 21 10.5

S S 9 Two Solar Panels Time-multiplexed 22 10.6


flash ROM inside the UR-SolarCap board, when the su-percapacitor block voltage is higher than 3.5 V. Below 3.5 V, the SEPIC regulator that provides 5 V to the mi-crocontroller is not functional, thereby preventing both the microcontroller and its programming capabilities to be operational.

Data Logging: UR-SolarCap measures the input volt-age and input current of its power source, as well as the supercapacitor block voltage and transmits this infor-mation through its RS-232 port. Although the original data transmission period is set to three minutes in the UR- SolarCap firmware, we modified the firmware to de-crease this period to 7 ms for finer temporal data resolu-tion in our experimental investigations. We also included the instantaneous SEPIC duty cycle value D in the trans-mitted information.

The logged data by the UR-SolarCap boards is used to analyze the system performance in all of the experi-ments except for the wakeup functionality (Section 10.2) and the S S configuration (Section 10.6). For the S S experiment (Section 10.6), because we needed a sam-pling frequency that is much higher than the capability of the built-in PIC ADC in UR-SolarCap, we used an Agilent MSO-X 2024A oscilloscope to log solar panel voltages. In the wakeup functionality experiment (Sec-tion 10.2), because the system starts from a totally depleted energy state and the logging feature of UR-SolarCap is not available, we used Agilent 1272 DMMs (sampling rate of 1 second).

10.2. Wakeup Operation with a Wind TurbineThe original UR-SolarCap board [34] is a solar-only har-vester that incorporates auto wake-up functionality, which allows it to resume normal operation from a fully depleted energy buffer state. In Section 6.2, we presented the necessary hardware and software modifications to

turn UR-SolarCap into a W harvester and confirmed its general harvesting functionality (Fig. 11). In this section, we provide experimental results validating the wake-up functionality of our W harvester. In this experiment, a fully-depleted 8 × 3000 F supercapacitor block was used as the energy buffer. The wind turbine voltage, microcon-troller supply voltage, and the supercapacitor block volt-age were logged. The results are shown in Fig. 17. When the supercapacitor voltage is less than 3.5 V and the input voltage is less than 6 V, the system is completely passive with the PIC microcontroller nonfunctional. As the input voltage rises above 6 V, wind power is supplied to the system through the wakeup path and the supercapacitor voltage starts increasing. Once the supercapacitor block exceeds 3.5 V, the system wakes up, the microcontroller becomes operational and starts intelligent harvesting, including MPPT. This can be seen in the figure as the constant 5 V supply voltage of the microcontroller in UR-SolarCap after the first hour.

10.3. Independent Solar/Wind Energy HarvestingFigure 18 demonstrates the harvesting performance of the SW configuration (Section 7.1). This setup includes two UR-SolarCap boards, a solar panel, and a wind tur-bine. The instantaneous power of each source was mea-sured through the measurement and communication module of UR-SolarCap [34]. Figure 18 clearly demon-strates the drastic differences between the nature of so-lar and wind power sources; while the power output of the solar panel exhibits gradual changes over time, the wind turbine supplies its power in short bursts peaking up to 30 W. The short power interruptions in Fig. 18, ap-proximately every four minutes, are software induced by the fractional MPPT algorithm. The algorithm dis-connects the solar panel from the harvester for these periods in order to measure the open-circuit voltage

25

20

15

Win

d T

urbi

ne V

olta

ge (

V)

10

5

0 1 2Time (h)

3

6

4

Mic

roco

ntro

ller

Vol

tage

(V

)

2

00 1 2

Time (h)3

6

4

Sup

erca

paci

tor

Vol

tage

(V

)

2

00 1 2

Time (h)

(a) (b) (c)

3

Figure 17. Wakeup operation of the ur-solarcap when used as a wind energy harvester (W configuration introduced in section 6.2). When the turbine voltage exceeds 6 V, microcontroller supply voltage reaches 5 V and the system starts active harvesting. because the wind comes in short gusts, the microcontroller supply voltage frequently falls below 5 V in the first hour. When the supercapacitor voltage exceeds 3.5 V, the system wakes up and the microcontroller supply voltage stays at 5 V.


VOC^ h of the solar panel to set the (approximate) MPP operating point and no power is harvested over these short periods.

10.4. Cooperative Solar Energy HarvestingThe Cooperative mode discussed in Section 8 (and de-picted in Fig. 13(c)) assigns the responsibility of set-ting the SEPIC duty cycle (for board1 and board2) to the master board (board1). An alternative approach that we tested allows both boards to perform their VOC mea-surement and set their own duty cycle independently. However, this can cause an imbalance in their incom-

ing power, as depicted in Fig. 19, which plots the power inputs of board1 and board2 in the S S2 2 configuration.

Our experimental setup to produce the results in Fig. 19 included two UR-SolarCap boards operating in Cooperative mode and two 2:1 analog MUXes, which connect two parallel 30 W solar panels to each of the harvesting boards. We do not present the results for cooperative multi-board harvesters in Regular mode, since their operation is identical to independent multi-board configurations in this mode.

Theoretically, since both boards are connected to the same power source, the calculated MPP is expected

35

30

25

Win

d P

ower

(W

)

20

15

10

5

00 5 10 15

Time (min)

3.5

3

2.5

Sol

ar P

ower

(W

)

2

1.5

1

0.5

00 5 10 15

Time (min)

(a) (b)

Figure 18. Input power from two ur-solarcap harvesters in the SW configuration over durations spanning 15 min. although the wind power appears in short bursts (a), its average is approximately 1.2 W, which is comparable to the average solar power (right). the mPP is calculated approximately every four minutes in the solar harvester firmware, which is observed as the small interrup-tions in the solar power (b).

10

8

6

4

2

0

60

48

Nor

mal

ized

Dut

y C

ycle

(%

)

Nor

mal

ized

Dut

y C

ycle

(%

)

36

24

12

0

Boa

rd1

Pow

er (

W)

0 1 2Time (min)

10

8

6

4

2

0

60

48

36

24

12

0

Boa

rd2

Pow

er (

W)

0 1 2Time (min)

(a) (b)

Figure 19. experimental results for the S S2 2 configuration, operating in Cooperative mode; one of the solar power sources was harvested by both boards cooperatively, while the other one was disconnected. during this experiment, board1 and board2 calculated their own sePIc duty cycles independently, which ended up being slightly different; this led to an unbalanced power distribution, with board1 receiving 68% and board2 receiving 32% of the total available 8.35 W.


to be the same in both boards. However, due to the pres-ence of noise in measurements, the determined SEPIC duty cycle values can be slightly different, leading to an unbalanced power distribution between the boards. The average power generated by the shared solar panel was approximately 8.35 W during the experiment. board1 re-ceived 5.65 W at its input, while board2 received 2.70 W on average, resulting in a load distribution of 68%–32%. At t = 0.8 minutes, both boards change their duty cy-cle value independently trying to keep the solar panel at their internally calculated ,VMPP causing the abrupt changes to the solar power input. Note that the gener-ated solar power (8.35 W) is relatively low compared to the rated power of the solar panels (60 W), as the experi-ments were conducted during a cloudy day.

Figure 20 shows the operation of an S S2 2 setup in Co-operative mode, in which the master board determines and shares the value of the SEPIC duty cycle. This im-plementation is based on our description of the coop-eration mechanism in Section 8.2. As can be seen, the source power is evenly distributed between the master and slave boards, as they now use the same duty cycle value. During this experiment, the shared solar panel delivered . 6.91 W on average, 3.55 W was distributed to board1 and 3.36 W to board2 corresponding to a 51% and 49% load split.

10.5. Cooperative Wind Energy HarvestingFigure 21 depicts the operation of the cooperative multi-board configuration operating in Cooperative mode, where the master board calculates the SEPIC duty cycle and shares it with the slave. During this ex-periment, we connected a wind turbine to both of the

harvesting boards and programmed the firmware—of both boards—to stay in Cooperative mode, there-by jointly harvesting the single wind power source. The results of this experiment are applicable to the

, ,W W S W2 2 2 2 and W2 harvesters, operating in Coopera-tive mode as follows:

■ While both boards harvest a different wind pow-er source independently in the Regular mode of a W W2 2 harvester, they decide to switch to Coopera-tive mode when the power level of one of the sourc-es increases substantially; in Cooperative mode, the wind turbine with higher power is harvested by both boards, through a configuration that is identi-cal to our experimental setup;

■ In S W2 2 harvesters, solar and wind power sourc-es are independently harvested by two boards in Regular mode, while the wind power source is harvested cooperatively when its power level in-creases significantly, prompting a switch to the Cooperative mode; this is exactly the experimen-tal setup we have;

■ The operation of a W2 configuration is identical to our experimental setup already in Coopera-tive mode.

The results of this experiment are depicted in Fig. 21. During the experiment, board1 and board2 received an average power of 0.86 W and 0.88 W, respectively, which represents a load balance of 49% and 51%.

10.6. Time-multiplexed Energy HarvestingTo analyze the behavior of our time-multiplexed single-board architecture, we conducted an experiment on our S S configuration (Section 10.6 and Fig. 14), in

5.5

5

4.5

4

3

3.5

2.5

5.5

5

4.5

4

3

3.5

2.5

60

50

Nor

mal

ized

Dut

y C

ycle

(%

)

Nor

mal

ized

Dut

y C

ycle

(%

)

40

30

10

20

0

60

50

40

30

10

20

0

Boa

rd1

Pow

er (

W)

0 1 32Time (min)

Boa

rd2

Pow

er (

W)

0 1 32Time (min)

(a) (b)

Figure 20. two S systems in the S S2 2 configuration of cooperative multi-board solar energy harvesting architecture operating in Cooperative mode. In this mode, board1 sends the sePIc duty cycle value to board2. as we can see, the total solar power is approximately distributed evenly between the boards.


which two separate solar panels—with a rated power of 30 W—were connected to a single harvesting board through a 2:1 analog MUX. We programmed the harvest-ing board to keep the voltage of one solar panel at 12 V and the other at 13.8 V, thereby, simulating a situation where two panels are at different locations/orientations and have different VMPP voltages. Considering the out-put power of solar panels (6 W and 15 W), we set the multiplexer round robin sequencing period to 3 ms and used a 100 nF input buffer capacitor for the UR-SolarCap board, which is connected to the output of the MUX. We used large 4700 nF capacitors at each input of the MUX for both power sources. This eliminates the situation

where each power source has a significant ripple during the switches.

Figure 22 depicts the results of our experiment. As discussed in Section 10.6, reducing the ripples around VMPP is critical in time-multiplexed single board configu-rations to ensure high system efficiency (particularly due to .)swh During this experiment, harvesting board was able to keep the panels’ average voltage within ±5% of the programmed VMPP values.

The total round-robin sequencing period of the MUX was 6 ms, during which each source was harvested for 3 ms. During the first 3 ms, the firmware adjusted the SEPIC duty cycle of the switching signal to ,D1 which

15

14

13

12

11

Inpu

t Sou

rce

Vol

tage

(V

)

0 50 200150100Time (ms)

15

14

13

12

11

Inpu

t Sou

rce

Vol

tage

(V

)

0 50 200150100Time (ms)

(a) (b)

Figure 22. experimental results for the SS configuration of time-multiplexed single-board architecture. the firmware was modi-fied to keep the first solar panel at 12 V (6 W) and the second solar panel at 13.8 V (15 W) to emulate the situation where the two solar panels have different mPP voltages due to being placed at different physical locations and orientations. the single harvest-ing board was able to simultaneously keep the two solar panels within 5% of their programmed mPP voltage value.

30

25

20

15

5

10

0

30

25

20

15

5

10

0

100

0

Nor

mal

ized

Dut

y C

ycle

(%

)

100

0

Nor

mal

ized

Dut

y C

ycle

(%

)

Boa

rd1

Pow

er (

W)

0 2 864Time (min)

0 2 864Time (min)

(a) (b)

Boa

rd2

Pow

er (

W)

Figure 21. experimental results for the cooperative multi-board harvesting configuration operating in Cooperative mode (intro-duced in section 8), where a single wind power source is commonly harvested by two ur-solarcap boards, while the other power source is left disconnected. as the type of the disconnected source does not affect the harvesting performance, these results are applicable to the ,W2 ,W W2 2 and S W2 2 configurations.


kept the first source at V 121MPP = V; upon switching to the other source, the SEPIC duty cycle was instantly changed to D2 to keep the second source at .V 13 82MPP = V during the second 3 ms interval. At the end of each 3 ms interval, the value of the SEPIC duty cycle was stored as the starting point for when the round-robin sequence selected this source again.

Which specialist should you see for diagnosing a spi-nal cord injury which specialist should you see for diag-nosing a spinal cord injury

11. Conclusions and Future WorkIn this paper, we reviewed solar and wind power har-vesting, highlighted the utility of using these sources as a pair of complementary power sources, and presented several hybrid harvester architectures based on open-

source designs that combine solar/wind harvesting. Our analysis confirms that hybrid harvesters for solar/wind power sources allow embedded systems to be deployed with much less downtime, as compared to systems us-ing solar or wind power alone. We provide four differ-ent classes of energy harvester system designs to cover a wide range of trade-offs between cost, complexity, fault-tolerance, and expandability. Table 10 lists the four categories of our designs, which are available as open-source designs from http://www.ece.rochester.edu/ projects/siplab/OpenWare/UR-SolarCap.html, along with the designs for UR-SolarCap—the open-source solar-only harvester on which these designs are based.

Our first category is single-source harvesters, for which we introduce the notations S and W to denote solar only and wind only, respectively. While the S

Table 10. A comparison among all presented configurations and their advantages/disadvantages, annotated using the symbols / and ,0 respectively. The choice of the best setting should be made according to the system’s applications and availability of power sources. Bullet points ( • ) under the “Sources” column represent the number and type of the power sources in each configuration. For example, the SS W configuration includes two solar panels and one wind turbine.

Architecture Notation Sources Advantages/Disadvantages

Solar Wind

Single-Source Single-Board S • / Simplicity 0 Not HybridW •

Independent Multiple Board SW • • / Simplicity

/ Expandability

/ Fault-Tolerance

0 Cost

SS ••WW ••SSS •••WWW •••SSW •• •

Cooperative Multiple Board S2 • / Increased Pceil

/ Fault-Tolerance

0 Simplicity

0 Expandability

0 Cost

W2 •

S S2 2 ••

W W2 2 ••

S W2 2 • •

Time Multiplexed Single Board SS •• / Cost

/ Expandability

0 Fault-Tolerance

0 Simplicity

W W ••

S W • •

S S S •••

W W W •••

S WW • ••

S WS •• •


configuration is based on the original UR-SolarCap de-sign, the W configuration is based on modified firmware for the UR-SolarCap board introduced in this paper.

In our second independent multiple-board hybrid har-vester category, multiple UR-SolarCap boards harvest a single power source independently and buffer their output into the same supercapacitor block; we use the notations SW, SS, WW to denote hybrid power sources with two power inputs, solar/wind, solar/solar, and wind/wind, respectively. The SSS, WWW, SSW, and SWW no-tations denote three independent power sources, solar/solar/solar, f , solar/wind/wind. Despite its cost disad-vantage, this category is highly expandable, because another power source can be very easily added by add-ing another UR-SolarCap board with its input connected to the new power source and its output to the common supercapacitor block. Furthermore, this category pro-vides fault resilience against the malfunction of UR- SolarCap boards.

Our next hybrid harvester category is termed coop-erative multiple-board and uses multiple UR-SolarCap boards. In contrast with the previous category, this new category allows multiplexing of different power sources into multiple UR-SolarCap boards. Such multiplexing in-troduces an advantage: if the power level of a specific power source is too high for a single UR-SolarCap board to handle at any given point in time (defined as eilPC in Eq. 25) and the power input of the other power source is fairly low, two UR-SolarCap boards can be used to harvest the same input, potentially doubling the har-vested power. We introduce a subscript notation (··2 ) to denote the number of UR-SolarCap boards, harvest-ing cooperatively; for example the S2 and W2 notations denote the configuration with two UR-SolarCap boards, with either a single solar ( S2 ) or a single wind (W2 ) in-put, thereby permitting doubled harvesting capability for either power input. Similarly the , ,S S W W2 2 2 2 and S W2 2 notations denote two separate solar power inputs ( S S2 2 ), two separate wind power inputs (W W2 2 ), and one wind and one solar power input (S W2 2 ) being har-vested cooperatively by two UR-SolarCap boards. Three power-input configurations of the cooperative harvest-ing, ,S S W3 3 3 and ,S W W3 3 3 denote the usage of three UR-SolarCap boards cooperatively harvesting solar/so-lar/wind and solar/wind/wind power inputs, the former being a good candidate for hybrid harvesting that uses two separate solar panels, facing the sun at two differ-ent angles to achieve a more steady solar power input throughout the day.

Our final hybrid harvester category is termed time-multiplexed single-board, which uses only a single UR-SolarCap board to harvest energy available at multiple power sources. This configuration uses time-multiplex-

ing to harvest each individual power source for only a portion of the time, while buffering the other power sources using small capacitors. This configuration has a cost advantage and lends itself well to expansion; how-ever, it suffers from reduced fault tolerance, because the single UR-SolarCap board used in this configuration cre-ates a single point of failure for all of the harvested pow-er sources. We use the notations , ,S SS W and W W to denote a configuration that uses a single UR-SolarCap board, which time-multiplexes two power inputs, so-lar/solar, solar/wind, and wind/wind, respectively. The three-power-input versions of the same configuration are ,S S S ,W W W W WS and .SS W

To test our designs, we conduct experiments on se-lected configurations of each architecture and evaluate their functionality. Our experimental results confirm the efficacy and versatility of our proposed system architec-tures and also highlight the advantages and disadvan-tages of each design, as listed in Table 10. In future work, we plan to introduce versions of our designs that are ca-pable of harvesting a variety of other power sources, in addition to solar and wind.

AcknowledgmentThis work was supported in part by the National Science Foundation grant CNS-1239423.

Mohamadhadi Habibzadeh received his B.S. in Computer Engineering from Isfahan University of Technology in Iran in 2015 and his M.S. degree in Technical Entrepreneurship and Management (TEM) from University of -Rochester,

USA in 2016. He is currently pursing his PhD degree in the Electrical and Computer Engineering department of Uni-versity at Albany, State University of New York (SUNY Al-bany), under the supervision of Dr. Tolga Soyata. His cur-rent research interests include Cyber Physical systems and embedded systems with applications in Internet of Things and Smart Cities.

Moeen Hassanalieragh received his B.S. degree in Electrical Engineering from Sharif University of Technology, Tehran, Iran in 2012. He joined the Uni-versity of Rochester Electrical and Computer Engineering Department (UR

ECE) in January 2014. He received his M.S. degree from UR ECE in 2015. He has been pursuing his PhD degree at UR ECE; until mid-2016, he was with Dr. Soyata’s research group. His research interests includes energy aware computing, energy harvesting circuits, Cyber Physical Systems, and Terahertz Imaging.


Akihiro Ishikawa was an undergradu-ate student at the University of Roches-ter, ECE department until he received his BS ECE degree in 2016. He worked as an undergraduate research assistant, helping Dr. Sharma and Dr. Soyata with

research in energy harvesting systems. He is currently pursuing his MS degree at Carnegie Mellon University.

Tolga Soyata (M’08 SM’16) received his B.S. degree in Electrical and Communi-cations Engineering from Istanbul Tech-nical University in 1988, M.S. degree in Electrical and Computer Engineering from Johns Hopkins University in 1992

and Ph.D. in Electrical and Computer Engineering from University of Rochester in 2000. He joined the University of Rochester ECE Department in 2008. He was an Assis-tant Professor—Research at UR ECE when he left to join SUNY Albany Department of ECE as an Associate Profes-sor in 2016. His teaching interests include CMOS VLSI ASIC Design, FPGA-based High Performance Data Pro-cessing System Design, and GPU Architecture and Paral-lel Programming. His research interests include Cyber Physical Systems, Digital Health, and GPU-based high-performance computing. He is a senior member of both IEEE and ACM.

Gaurav Sharma (S’88-M’96-SM’00-F’13) is a professor at the University of Roch-ester in the Department of Electrical and Computer Engineering, in the Depart-ment of Computer Science and in the De-partment of Biostatistics and Computa-

tional Biology. He received the BE degree in Electronics and Communication Engineering from Indian Institute of Technology Roorkee (formerly Univ. of Roorkee), India in 1990; the ME degree in Electrical Communication Engi-neering from the Indian Institute of Science, Bangalore, India in 1992; and the MS degree in Applied Mathematics and PhD degree in Electrical and Computer Engineering from North Carolina State University, Raleigh in 1995 and 1996, respectively. From Aug. 1996 through Aug. 2003, he was with Xerox Research and Technology, in Webster, NY, initially as a Member of Research Staff and subsequently at the position of Principal Scientist. Dr. Sharma’s re-search interests include data analytics, cyberphysical systems, signal and image processing, computer vision, media security, and communications. He is a fellow of the IEEE, of SPIE, and of the Society of Imaging Science and Technology (IS&T) and a member of Sigma Xi. From 2011 through 2015, he served as the Editor-in-Chief for the Journal of Electronic Imaging and in the past has served

as an associate editor for the Journal of Electronic Imag-ing, IEEE Transactions on Image Processing, and IEEE Transactions on Information Forensics and Security.

References[1] B. Morris and M. Trivedi, “Real-time video based highway traffic measurement and performance monitoring,” in Proc. IEEE Intelligent Transportation Systems Conf., Sept. 2007, pp. 59–64.[2] F. Qu, F. Y. Wang, and L. Yang, “Intelligent transportation spaces: Vehicles, traffic, communications, and beyond,” IEEE Commun. Mag., vol. 48, no. 11, pp. 136–142, Nov. 2010.[3] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet of things for smart cities,” IEEE Internet Things J., vol. 1, no. 1, pp. 22–32, Feb. 2014.[4] M. Habibzadeh, Z. Qin, T. Soyata, and B. Kantarci, “Large scale dis-tributed dedicated- and non-dedicated smart city sensing systems,” IEEE Sensors J., to be published.[5] V. Dyo, S. A. Ellwood, D. W. Macdonald, A. Markham, N. Trigoni, R. Wohlers, C. Mascolo, B. Pásztor, S. Scellato, and K. Yousef, “Wildsens-ing: Design and deployment of a sustainable sensor network for wildlife monitoring,” ACM Trans. Senors. Netw., vol. 8, no. 4, pp. 29:1–29:33, Sept. 2012.[6] F. Mehdipour, “Smart field monitoring: An application of cyber-phys-ical systems in agriculture (work in progress),” in Proc. IIAI 3rd Int. Conf. Advanced Applied Informatics, Aug. 2014, pp. 181–184.[7] J. Burrell, T. Brooke, and R. Beckwith, “Vineyard computing: Sensor networks in agricultural production,” IEEE Pervasive Comput., vol. 3, no. 1, pp. 38–45, Jan. 2004.[8] J. McCulloch, P. McCarthy, S. M. Guru, W. Peng, D. Hugo, and A. Ter-horst, “Wireless sensor network deployment for water use efficiency in irrigation,” in Proc. Workshop on Real-world Wireless Sensor Networks, New York, 2008, pp. 46–50.[9] M. Habibzadeh, W. Xiong, M. Zheleva, E. K. Stern, B. H. Nussbaum, and T. Soyata, “Smart city sensing and communication sub-infrastruc-ture,” in IEEE Midwest Symp. Circuits and Systems, Boston, MA, Aug. 2017.[10] Q. Zhou, A. Sussman, J. Chang, J. Dong, A. Zettl, and W. Mickelson, “Fast response integrated MEMS microheaters for ultra low power gas detection,” Sensors Actuators A: Phys., vol. 223, pp. 67–75, 2015.[11] Y. Lee, S. Bang, I. Lee, Y. Kim, G. Kim, M. H. Ghaed, P. Pannuto, P. Dutta, D. Sylvester, and D. Blaauw, “A modular 1 mm3 die-stacked sens-ing platform with low power I2C inter-die communication and multi-modal energy harvesting,” IEEE J. Solid-State Circuits, vol. 48, no. 1, pp. 229–243, Jan. 2013. [12] F. Dressler, S. Ripperger, M. Hierold, T. Nowak, C. Eibel, B. Cassens, F. Mayer, K. Meyer-Wegener, and A. Kolpin, “From radio telemetry to ultra-low-power sensor networks: Tracking bats in the wild,” IEEE Com-mun. Mag., vol. 54, no. 1, pp. 129–135, Jan. 2016.[13] M. H. Ghaed, G. Chen, R. u. Haque, M. Wieckowski, Y. Kim, G. Kim, Y. Lee, I. Lee, D. Fick, D. Kim, M. Seok, K. D. Wise, D. Blaauw, and D. Sylvester, “Circuits for a cubic-millimeter energy-autonomous wireless intraocular pressure monitor,” IEEE Trans. Circuits Syst. I, vol. 60, no. 12, pp. 3152–3162, Dec. 2013.[14] S. Nawaz and C. Mascolo, “Mining users’ significant driving routes with low-power sensors,” in Proc. 12th ACM Conf. Embedded Network Sensor Systems, New York, 2014, pp. 236–250.[15] A. Dementyev and J. A. Paradiso, “Wristflex: Low-power gesture in-put with wrist-worn pressure sensors,” in Proc. 27th Annual ACM Symp. User Interface Software and Technology, New York, 2014, pp. 161–166. [16] T. Torfs, T. Sterken, S. Brebels, J. Santana, R. van den Hoven, V. Spiering, N. Bertsch, D. Trapani, and D. Zonta, “Low power wireless sen-sor network for building monitoring,” IEEE Sensors J., vol. 13, no. 3, pp. 909–915, Mar. 2013.[17] A. Engel and A. Koch, “Heterogeneous wireless sensor nodes that target the internet of things,” IEEE Micro, vol. 36, no. 6, pp. 8–15, Nov. 2016.[18] A. Somov, A. Baranov, D. Spirjakin, and R. Passerone, “Circuit de-sign and power consumption analysis of wireless gas sensor nodes: One-sensor versus two-sensor approach,” IEEE Sensors J., vol. 14, no. 6, pp. 2056–2063, June 2014.[19] A. D. S. Bernabe, J. R. M. de Dios, and A. Ollero, “Efficient clus-ter-based tracking mechanisms for camera-based wireless sensor


networks,” IEEE Trans. Mobile Comput., vol. 14, no. 9, pp. 1820–1832, Sept. 2015.[20] M. Imran, K. Shahzad, N. Ahmad, M. O’Nils, N. Lawal, and B. Oel-mann, “Energy-efficient SRAM FPGA-based wireless vision sensor node: SENTIOF-CAM,” IEEE Trans. Circuits Syst. Video Technol., vol. 24, no. 12, pp. 2132–2143, Dec. 2014.[21] N. Kong and D. S. Ha, “Low-power design of a self-powered piezoelectric energy harvesting system with maximum power point tracking,” IEEE Trans. Power Electron., vol. 27, no. 5, pp. 2298–2308, May 2012.[22] Y. K. Ramadass and A. P. Chandrakasan, “A battery-less thermo-electric energy harvesting interface circuit with 35 mV startup voltage,” IEEE J. Solid-State Circuits, vol. 46, no. 1, pp. 333–341, Jan. 2011.[23] A. Meehan, H. Gao, and Z. Lewandowski, “Energy harvesting with microbial fuel cell and power management system,” IEEE Trans. Power Electron., vol. 26, no. 1, pp. 176–181, Jan. 2011.[24] M. Piñuela, P. D. Mitcheson, and S. Lucyszyn, “Ambient RF energy harvesting in urban and semi-urban environments,” IEEE Trans. Micro-wave Theory Tech., vol. 61, no. 7, pp. 2715–2726, July 2013.[25] D. Ramasur and G. P. Hancke, “A wind energy harvester for low power wireless sensor networks,” in Proc. IEEE Int. Instrumentation and Measurement Technology Conf., May 2012, pp. 2623–2627.[26] C. Alippi and C. Galperti, “An adaptive system for optimal solar en-ergy harvesting in wireless sensor network nodes,” IEEE Trans. Circuits Syst., vol. 55, no. 6, pp. 1742–1750, July 2008.[27] C. Park and P. H. Chou, “Power utility maximization for multiple-supply systems by a load-matching switch,” in Proc. Int. Symp. Low Power Electronics and Design, Aug. 2004, pp. 168–173.[28] K. Y. Lo, Y. M. Chen, and Y. R. Chang, “MPPT battery charger for stand-alone wind power system,” IEEE Trans. Power Electron., vol. 26, no. 6, pp. 1631–1638, June 2011.[29] C. Liu, K. T. Chau, and X. Zhang, “An efficient wind photovoltaic hy-brid generation system using doubly excited permanent-magnet brush-less machine,” IEEE Trans. Ind. Electron., vol. 57, no. 3, pp. 831–839, Mar. 2010.[30] T. Soyata, L. Copeland, and W. Heinzelman, “RF energy harvesting for embedded systems: A survey of tradeoffs and methodology,” IEEE Circuits Syst. Mag., vol. 16, no. 1, pp. 22–57, Feb. 2016.[31] S. J. Roundy, “Energy scavenging for wireless sensor nodes with a focus on vibration to electricity conversion,” Ph.D. dissertation, Univ. California, Berkeley, 2003.[32] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low level vibra-tions as a power source for wireless sensor nodes,” Computer Commun., vol. 26, no. 11, pp. 1131–1144, 2003.[33] National Renewable Energy Laboratory. [Online]. Available: http://rredc.nrel.gov/wind/pubs/atlas/tables/1-2T.html[34] M. Hassanalieragh, T. Soyata, A. Nadeau, and G. Sharma, “UR-Solar-Cap: An open source intelligent auto-wakeup solar energy harvesting system for supercapacitor based energy buffering,” IEEE Access, vol. 4, pp. 542–557, Mar. 2016.[35] X. Lu and S.-H. Yang, “Thermal energy harvesting for WSNs,” in Proc. IEEE Int. Conf. Systems, Man and Cybernetics, Oct. 2010, pp. 3045–3052.[36] R. Amirtharajah and A. P. Chandrakasan, “Self-powered signal processing using vibration-based power generation,” IEEE J. Solid-State Circuits, vol. 33, no. 5, pp. 687–695, 1998.[37] S. D. Moss, O. R. Payne, G. A. Hart, and C. Ung, “Scaling and power density metrics of electromagnetic vibration energy harvesting devic-es,” Smart Mater. Struct., vol. 24, no. 2, pp. 023001, 2015.[38] X. Xing, J. Lou, G. Yang, O. Obi, C. Driscoll, and N. Sun, “Wideband vibration energy harvester with high permeability magnetic material,” Appl. Phys. Lett., vol. 95, no. 13, pp. 134103, 2009.[39] D. L. Churchill, M. J. Hamel, C. P. Townsend, and S. W. Arms, “Strain energy harvesting for wireless sensor networks,” Smart Struct. Mater., vol. 5055, pp. 319–327, 2003.[40] S. Boisseau, G. Despesse, and A. Sylvestre, “Optimization of an electret-based energy harvester,” Smart Mater. Struct., vol. 19, no. 7, pp. 075015, 2010.[41] A. Dewan, H. Beyenal, and Z. Lewandowski, “Scaling up microbial fuel cells,” Environ. Sci. Technol., vol. 42, no. 20, pp. 7643–7648, 2008.[42] B. R. Ringeisen, E. Henderson, P. K. Wu, J. Pietron, R. Ray, B. Little, J. C. Biffinger, and J. M. Jones-Meehan, “High power density from a min-iature microbial fuel cell using shewanella oneidensis DSP10,” Environ. Sci. Technol., vol. 40, no. 8, pp. 2629–2634, 2006.

[43] Y. K. Tan and S. K. Panda, “Energy harvesting from hybrid indoor ambient light and thermal energy sources for enhanced performance of wireless sensor nodes,” IEEE Trans. Ind. Electron., vol. 58, no. 9, pp. 4424–4435, 2011.[44] G. Honan, N. Gekakis, M. Hassanalieragh, A. Nadeau, G. Sharma, and T. Soyata, “Energy harvesting and buffering for cyber physical sys-tems: A review,” in Cyber Physical Systems: A Computational Perspective. Boca Raton, FL: CRC, 2015, ch. 7, pp. 191–218.[45] K. M. Farinholt, G. Park, and C. R. Farrar, “RF energy transmission for a low-power wireless impedance sensor node,” IEEE Sensors J., vol. 9, no. 7, pp. 793–800, 2009.[46] B. Li, J. H. You, and Y.-J. Kim, “Low frequency acoustic energy har-vesting using PZT piezoelectric plates in a straight tube resonator,” Smart Mater. Struct., vol. 22, no. 5, pp. 055013, 2013.[47] X. Jiang, J. Polastre, and D. Culler, “Perpetual environmentally pow-ered sensor networks,” in Proc. Fourth Int. Symp. Information Processing in Sensor Networks, Apr. 2005, pp. 463–468.[48] E. O. Torres and G. A. Rincon-Mora, “Electrostatic energy-harvest-ing and battery-charging CMOS system prototype,” IEEE Trans. Circuits Syst., vol. 56, no. 9, pp. 1938–1948, Sept. 2009.[49] M. Zhu, M. Hassanalieragh, A. Fahad, Z. Chen, T. Soyata, and K. Shen, “Supercapacitor energy buffering for self-sustainable, continu-ous sensing systems,” Dept. Computer Sci., Univ. Rochester, Tech. Rep. TR–995, Mar. 2016.[50] M. Hassanalieragh, T. Soyata, A. Nadeau, and G. Sharma, “Solar-su-percapacitor harvesting system design for energy-aware applications,” in Proc. 27th IEEE Int. System-on-Chip Conf., Las Vegas, NV, Sept. 2014, pp. 280–285.[51] A. Fahad, T. Soyata, T. Wang, G. Sharma, W. Heinzelman, and K. Shen, “SOLARCAP: Super capacitor buffering of solar energy for self-sustainable field systems,” in Proc. 25th IEEE Int. System-on-Chip Conf., Niagara Falls, NY, Sept. 2012, pp. 236–241.[52] A. Nadeau, G. Sharma, and T. Soyata, “State-of-charge estimation for supercapacitors: A kalman filtering formulation,” in Proc. IEEE Int. Conf. Acoustics, Speech and Signal Processing, Florence, Italy, May 2014, pp. 2194–2198.[53] M. Habibzadeh, M. Hassanalieragh, T. Soyata, and G. Sharma, “Supercapacitor-based embedded hybrid solar/wind harvesting sys-tem architectures,” in Proc. 30th IEEE Int. System-on-Chip Conf., Munich, Germany, Sept. 2017.[54] A. S. Weddell, M. Magno, G. V. Merrett, D. Brunelli, B. M. Al-Hashimi, and L. Benini, “A survey of multi-source energy harvesting systems,” in Proc. Design, Automation Test in Europe Conf., Mar. 2013, pp. 905–908.[55] D. Carli, D. Brunelli, L. Benini, and M. Ruggeri, “An effective multi-source energy harvester for low power applications,” in Proc. Design, Automation Test in Europe, Mar. 2011, pp. 1–6.[56] C. Park and P. H. Chou, “AmbiMax: Autonomous energy harvest-ing platform for multi-supply wireless sensor nodes,” in Proc. 3rd Annu. IEEE Communications Society on Sensor and Ad Hoc Communications and Networks Conf., Sept. 2006, vol. 1, pp. 168–177.[57] M. Habibzadeh, M. Hassanalieragh, T. Soyata, and G. Sharma, “So-lar/wind hybrid energy harvesting for supercapacitor-based embedded systems,” in Proc. IEEE Midwest Symp. Circuits and Systems, Boston, MA, Aug. 2017.[58] A. Du Pasquier, I. Plitz, S. Menocal, and G. Amatucci, “A compara-tive study of li-ion battery, supercapacitor and nonaqueous asymmet-ric hybrid devices for automotive applications,” J. Power Sources, vol. 115, no. 1, pp. 171–178, 2003.[59] A. Nadeau, M. Hassanalieragh, G. Sharma, and T. Soyata, “Ener-gy awareness for supercapacitors using kalman filter state-of-charge tracking,” J. Power Sources, vol. 296, pp. 383–391, Nov. 2015.[60] S. Wilcox. (2012, Aug.). National solar radiation database 1991–2010 update, National Renewable Energy Laboratory, US Department of Energy, Tech. Rep. [Online]. Available: http://rredc.nrel.gov/solar/old_data/nsrdb/1991-2010/[61] A. Hande, T. Polk, W. Walker, and D. Bhatia, “Indoor solar energy harvesting for sensor network router nodes,” Microprocessors Micro-syst., vol. 31, no. 6, pp. 420–432, 2007.[62] A. Nasiri, S. A. Zabalawi, and G. Mandic, “Indoor power harvesting using photovoltaic cells for low-power applications,” IEEE Trans. Ind. Electron., vol. 56, no. 11, pp. 4502–4509, Nov. 2009.[63] J. A. Paradiso and T. Starner, “Energy scavenging for mobile and wireless electronics,” IEEE Pervasive Comput., vol. 4, no. 1, pp. 18–27, Jan. 2005.


[64] B. Anspaugh, GaAs Solar Cell Radiation Handbook, 1996.[65] J. Britt and C. Ferekides, “Thin-film CdS/CdTe solar cell with 15.8% efficiency,” Appl. Phys. Lett., vol. 62, no. 22, pp. 2851–2852, 1993.[66] A. Rockett and R. Birkmire, “CuInSe2 for photovoltaic applica-tions,” J. Appl. Phys., vol. 70, no. 7, pp. R81–R97, 1991.[67] P. Würfel and U. Würfel, Physics of Solar Cells: From Basic Principles to Advanced Concepts. New York: Wiley, 2009.[68] G. Vachtsevanos and K. Kalaitzakis, “A hybrid photovoltaic simula-tor for utility interactive studies,” IEEE Trans. Energy Conversion, vol. EC-2, no. 2, pp. 227–231, June 1987.[69] K. H. Hussein, I. Muta, T. Hoshino, and M. Osakada, “Maximum photovoltaic power tracking: An algorithm for rapidly changing atmo-spheric conditions,” IEEE Proc. Generation Transmission Distribut., vol. 142, no. 1, pp. 59–64, Jan. 1995.[70] T. Esram and P. L. Chapman, “Comparison of photovoltaic array maximum power point tracking techniques,” IEEE Trans. Energy Conver-sion, vol. 22, no. 2, pp. 439–449, June 2007.[71] J. J. Schoeman and J. D. V. Wyk, “A simplified maximal power con-troller for terrestrial photovoltaic panel arrays,” in Proc. IEEE Power Electronics Specialists Conf., June 1982, pp. 361–367.[72] W. Xiao and W. G. Dunford, “A modified adaptive hill climbing MPPT method for photovoltaic power systems,” in Proc. IEEE 35th Annu. Power Electronics Specialists Conf., June 2004, vol. 3, pp. 1957–1963. [73] C. Hua, J. Lin, and C. Shen, “Implementation of a DSP-controlled photovoltaic system with peak power tracking,” IEEE Trans. Ind. Elec-tron., vol. 45, no. 1, pp. 99–107, Feb. 1998.[74] P. Midya, P. T. Krein, R. J. Turnbull, R. Reppa, and J. Kimball, “Dy-namic maximum power point tracker for photovoltaic applications,” in Proc. 27th Annu. IEEE Power Electronics Specialists Conf., June 1996, vol. 2, pp. 1710–1716.[75] T. Khatib, A. Mohamed, N. Amin, and K. Sopian, “An efficient maxi-mum power point tracking controller for photovoltaic systems using new boost converter design and improved control algorithm,” WSEAS Trans. Power Syst., vol. 5, no. 2, pp. 53–63, 2010.[76] E. Koutroulis and K. Kalaitzakis, “Design of a maximum power tracking system for wind-energy-conversion applications,” IEEE Trans. Ind. Electron., vol. 53, no. 2, pp. 486–494, Apr. 2006.[77] T. Thiringer and J. Linders, “Control by variable rotor speed of a fixed-pitch wind turbine operating in a wide speed range,” IEEE Trans. Energy Conversion, vol. 8, no. 3, pp. 520–526, Sept. 1993.[78] S. M. Barakati, M. Kazerani, and X. Chen, “A new wind turbine gen-eration system based on matrix converter,” in Proc. IEEE Power Engi-neering Society General Meeting, June 2005, vol. 3, pp. 2083–2089.[79] W. Lu and B. T. Ooi, “Multiterminal LVDC system for optimal acqui-sition of power in wind-farm using induction generators,” IEEE Trans. Power Electron., vol. 17, no. 4, pp. 558–563, July 2002.[80] S. M. Barakati, M. Kazerani, and J. D. Aplevich, “Maximum power tracking control for a wind turbine system including a matrix con-verter,” IEEE Trans. Energy Conversion, vol. 24, no. 3, pp. 705–713, Sept. 2009.[81] S. Chapman, Electric Machinery Fundamentals. Tata McGraw-Hill Education, 2005.[82] K. Amei, Y. Takayasu, T. Ohji, and M. Sakui, “A maximum power con-trol of wind generator system using a permanent magnet synchronous generator and a boost chopper circuit,” in Proc. IEEE Power Conversion Conf., vol. 3., 2002, pp. 1447–1452.[83] National Oceanic and Atmospheric Administration. [Online]. Avail-able: www.noaa.gov[84] P. Giorsetto and K. F. Utsurogi, “Development of a new procedure for reliability modeling of wind turbine generators,” IEEE Trans. Power Apparatus Syst., vol. PAS-102, no. 1, pp. 134–143, Jan. 1983.[85] S. H. Jangamshetti and V. G. Rau, “Normalized power curves as a tool for identification of optimum wind turbine generator param-eters,” IEEE Trans. Energy Conversion, vol. 16, no. 3, pp. 283–288, Sept. 2001.[86] S. Chalasani and J. M. Conrad, “A survey of energy harvesting sources for embedded systems,” in Proc. IEEE Southeast Conf. Apr. 2008, pp. 442–447.[87] R. Morais, S. G. Matos, M. A. Fernandes, A. L. Valente, S. F. Soares, P. Ferreira, and M. Reis, “Sun, wind and water flow as energy supply for small stationary data acquisition platforms,” Computers Electron. Agri., vol. 64, no. 2, pp. 120–132, 2008.

[88] M. Hassanalieragh, A. Page, T. Soyata, G. Sharma, M. K. Aktas, G. Mateos, B. Kantarci, and S. Andreescu, “Health monitoring and man-agement using internet-of-things (IoT) sensing with cloud-based pro-cessing: Opportunities and challenges,” in Proc. IEEE Int. Conf. Services Computing, New York, June 2015, pp. 285–292.[89] A. Page, O. Kocabas, T. Soyata, M. K. Aktas, and J. Couderc, “Cloud-based privacy-preserving remote ECG monitoring and surveillance,” Ann. Noninvasive Electrocardiol., vol. 20, no. 4, pp. 328–337, 2014.[90] A. Page, S. Hijazi, D. Askan, B. Kantarci, and T. Soyata “Research directions in cloud-based decision support systems for health moni-toring using internet-of-things driven data acquisition,” Int. J. Services Comput., vol. 4, no. 4, pp. 18–34, 2016.[91] N. Gekakis, A. Nadeau, M. Hassanalieragh, Y. Chen, Z. Liu, G. Hon-an, F. Erdem, G. Sharma, and T. Soyata, “Modeling of supercapacitors as an energy buffer for cyber-physical systems,” in Cyber Physical Sys-tems: A Computational Perspective. Boca Raton, FL: CRC, 2015, ch. 6, pp. 171–190.[92] D. Raví, C. Wong, B. Lo, and G. Z. Yang, “A deep learning approach to on-node sensor data analytics for mobile or wearable devices,” IEEE J. Biomed. Health Inform., vol. 21, no. 1, pp. 56–64, Jan. 2017.[93] R. Pérez-Torres, C. Torres-Huitzil, and H. Galeana-Zapién, “Power management techniques in smartphone-based mobility sensing sys-tems: A survey,” Pervasive Mobile Comput., vol. 31, pp. 1–21, 2016.[94] F. I. Simjee and P. H. Chou, “Efficient charging of supercapacitors for extended lifetime of wireless sensor nodes,” IEEE Trans. Power Elec-tron., vol. 23, no. 3, pp. 1526–1536, May 2008.[95] Microchip PIC16F1789. [Online]. Available: http://www.microchip .com/wwwproducts/en/PIC16F1783[96] Designing DC/DC converters based on SEPIC topology, Texas In-struments [Online]. Available: http://www.ti.com/lit/an/slyt309/slyt309.pdf[97] O. Kocabas, T. Soyata, and M. K. Aktas, “Emerging security mecha-nisms for medical cyber physical systems,” IEEE/ACM Trans. Comput. Biol. Bioinform., vol. 13, no. 3, pp. 401–416, June 2016.[98] T. Soyata, Enabling Real-Time Mobile Cloud Computing through Emerging Technologies. IGI Global, 2015.[99] G. Honan, A. Page, O. Kocabas, T. Soyata, and B. Kantarci, “Inter-net-of-everything oriented implementation of secure digital health (D-Health) systems,” in Proc. IEEE Symp. Computers and Communica-tions, Messina, Italy, June 2016, pp. 718–725.[100] A. Page, M. Hassanalieragh, T. Soyata, M. K. Aktas, B. Kantarci, and S. Andreescu, “Conceptualizing a real-time remote cardiac health monitoring system,” in Enabling Real-Time Mobile Cloud Computing through Emerging Technologies, T. Soyata, Ed. IGI Global, 2015, ch. 1, pp. 1–34.[101] E. Koutroulis and K. Kalaitzakis, “Design of a maximum power tracking system for wind-energy-conversion applications,” IEEE Trans. Ind. Electron., vol. 53, no. 2, pp. 486–494, Apr. 2006.[102] S. M. R. Kazmi, H. Goto, H. J. Guo, and O. Ichinokura, “A novel al-gorithm for fast and efficient speed-sensorless maximum power point tracking in wind energy conversion systems,” IEEE Trans. Ind. Electron., vol. 58, no. 1, pp. 29–36, Jan. 2011.[103] Q. Wang and L. Chang, “An intelligent maximum power extraction algorithm for inverter-based variable speed wind turbine systems,” IEEE Trans. Power Electron., vol. 19, no. 5, pp. 1242–1249, Sept. 2004.[104] S. Morimoto, H. Nakayama, M. Sanada, and Y. Takeda, “Sensor-less output maximization control for variable-speed wind generation system using IPMSG,” IEEE Trans. Ind. Applicat., vol. 41, no. 1, pp. 60–67, Jan. 2005.[105] E. Koutroulis, K. Kalaitzakis, and N. C. Voulgaris, “Development of a microcontroller-based, photovoltaic maximum power point tracking control system,” IEEE Trans. Power Electron., vol. 16, no. 1, pp. 46–54, Jan. 2001.[106] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, “Optimization of perturb and observe maximum power point tracking method,” IEEE Trans. Power Electron., vol. 20, no. 4, pp. 963–973, July 2005.[107] GNU Octave. [Online]. Available: https://www.gnu.org/software/octave/about.html[108] H. Mousazadeh, A. Keyhani, A. Javadi, H. Mobli, K. Abrinia, and A. Sharifi, “A review of principle and sun-tracking methods for maximizing solar systems output,” Renewable Sustain. Energy Rev., vol. 13, no. 8, pp. 1800–1818, 2009.

hybrid solar-wind energy harvesting for embedded ...gsharma/papers/hadihybsolarwindh… · range...

Documents